Testing to Find the Fastest Bicycle Wheels

***Please note the graph and the data on this page will no longer be updated. The main issue being the control tyre is no longer manufactured by Continental. Additionally, test methods have moved on and Dnitriev number will be used as the basis for future testing***

Precursor – The critics and those with a vested interest

I have added this section to the start of this blog post. The method this test uses is called transient state and it is used when aerodynamics are constantly changing. This is difficult to explain in detail in one blog post so I have linked to a youtube video.

Forward to 10m20, this shows the clear difference between steady state and the real world

The important part of the video is at 10m20 and this shows the difference between airflow that exists in the real world, airflow that manufacturers test at and a (turbulent) transient test. The flow that is generated in a steady state wind tunnel rarely occurs in real life and is borderline acceptable for a velodrome.

Be aware that a lot of wheel companies use internet forum users as a method of subversively selling a reality that does not exist under a premise of an impossible power saving. I DO NOT SELL WHEELS SO I HAVE NO VESTED INTERESTS and remain impartial.

In short, the wheel companies are exaggerating their power savings and testing using methods THAT ARE INAPPROPRIATE FOR ROAD USE. If they tell you otherwise, they are categorically lieing and I would be more than happy to prove that in a court of law.


In terms of drag caused by a bicycle rider, the biggest loss is caused by the rider themselves followed by the wheels and frame.

The drag caused by wheels is significant because of two fundamental reasons. The first is they hit the air first as they are the most forward part of the bike and second because they are rotating. The effective air speed at the top of a wheel/tyre is double the indicated speed of the bike.

In the bike industry, wheel aerodynamic testing has generally been conducted by two groups of people – Wheel manufacturers and journalists. Wheel manufacturers will usually adjust tests to make their particular wheels look more favourable than their competitors in testing. This is usually achieved by a combination of adjusting speeds and angles. The reality is this type of test is not impartial.

Journalists on the other hand tend to visit their local university and ask some clever boffin to conduct their testing for them and give them the results or go to their local velodrome, hold a speed and see how much power the wheels consume.

Both of the above testing methodologies are not representative of the real world. A comparative analogy would be fuel consumption for a car driving along a perfectly smooth glass like road surface with no wind and no change in speed – it is totally unrealistic.

The testing that has been carried out is usually steady state. A steady state analysis assumes the wheels, bike and rider are in a nice environment where air is hitting them at a perfect speed and perfect angle. The drag is then recorded.

In the real world, very few riders have the ability to maintain a speed of 50km/h for a length of time as they are simply not fit enough. The reality is on the open road, wind does not come in from a perfect angle, it’s speed changes and things like street furniture (hedges, kerbs, passing cars, rider rocking from left to right) upset the airflow over the rider. Modelling this type of situation is called transient analysis. It is technically more difficult to carry out transient analysis both in CFD and in a wind tunnel. Most wind tunnels are not geared up to carry out transient analysis.

Wheel manufacturers are now using a weighted analysis of yaw angles and speeds to give an overall rating for their wheels. Bare in mind they can adjust their weighting to make their wheels look better!

A superior method of analysis is to carry out a transient analysis in a wind tunnel. This requires a wind tunnel with Horizontal and Vertical Louvres to add Swirl to the air before it hits the bike and rider. This allows a much more realistic estimate of drag to be estimated as it simulates road conditions.

General Guidance

Yaw Angles
Wheel manufacturers tout their wheels as having fantastic drag at varying yaw angles. The effectiveness of their marketing is remarkable as many posters on the Internet also believe this.

Due to the laws of physics, for an average rider, the maximum yaw angle before complete separation occurs is around 12 degrees. A more blunt (toroidal) cross section might get to 15 but that’s really the limit. This limit of separation is affected by a variable known as Reynolds number (A combination of Speed, density, profile of the shape and viscosity)

Aerodynamic design is always a compromise, increasing the separation point at high yaw angles will always negatively impact drag at very low (<5 degree) yaw angles.

In repeated testing, wheels with very good transient performance work best for the average rider.

This guidance is uniform across the board. It is vitally important to install tyres that are slightly narrower or inline with the brake track of the wheel rim. A ballooning tyre will impact the drag significantly.

There has been a trend towards wider tyres on bikes of late. From an aerodynamic perspective, the width of the rear tyre has little effect but the width of the front tyre has much more impact and therefore a 23mm front tyre is recommended irrespective of whether the wheel was designed for 25mm tyres. At speeds above 30km/h, it is more beneficial to have 23mm tyres than 25mm front tyres for aerodynamic benefit.

Testing Protocol

The test protocol is the product of “weekend work” by a group of Aerospace Engineers from Bristol, England. The testing protocol is very different to manufacturer tests. It is fundamentally impartial and mimics real world riding conditions in the sense it models transient air movement. Emphasis is placed on wheels which handle the separation and reattachment of airflow efficiently, very little emphasis is placed on riding a bike straight into a head wind at zero degree yaw – this is not realistic so why bother testing it. The wind tunnel used was temperature and humidity controlled.

The graph below shows a sample of one ride where a rider was riding along a straight road at an almost constant speed. It is clear that neither air speed or yaw angle were constant.

Road Test Data

The real world basis for this protocol are based on two subsets of bike riders in the Bristol (UK) area. Riders who are good club riders averaging 30km/h and time triallists averaging 50km/h. Data from their rides in terms of effective yaw angles, speed and air pressure distribution was recorded over 6 months. This was assessed, aggregated and mapped to a protocol suitable for a wind tunnel. The method of transformation was to statistically analyze the road conditions, apply a Fast Fourier Transformation to the data and run some test simulations for validation. The two discreet protocols are shown below.

THE GRAPHS DO NOT REPRESENT A RIDE CYCLE, THEY INDICATE THE PARAMETERS THE WHEELS WERE TESTED TO. Wind tunnels have limitations and part of the data gathering exercise is to validate data as it’s being processed. It is expensive and time consuming to rectify errors later. To replicate transient conditions, pulsing velocity or pulsing angle are both acceptable. The ramp tests were used to validate one against the other for each wheelset.

Test Protocol 30 Km/h
Test Protocol 50 Km/h

Through investigation it was found that micro corrections from the riders and the somewhat random nature of wind speed and deviation in yaw angle produced transient response of the bike and rider combination. This was much worse on the wheels as they were rotating into an oncoming airstream. In effect, a rider riding in a perfectly straight line into an oncoming wind was generating turbulence/buffeting/flutter by the bike rocking from side to side. What would be considered a zero degree yaw angle in steady state analysis behaves more like 5-6 degrees when the transient effects are factored in.

This protocol mimics the buffeting nature of the rider in the airstream configuration and produces an overall average drag value against time and consequently average power. It is designed to weed out the wheels that have poor transient performance. The lines on the protocols are shown for completeness, they do not mean this protocol favours blustery conditions.

Transient vs Steady State drag

The concept of transient drag effects have been well noted in low speed Aerospace applications such as military reconnaissance drones. This transient concept has not been applied to bicycle related products despite the overwhelming sensitivity of the velocity vectors involved. As an example the crosswind velocity on a bike often exceeds the forward velocity (Ratio > 1). A comparison for a car would yield a forward to crosswind ratio of 0.25 at 100km/h typical cruising speed.

A significant hurdle with trying to accurately measure the drag of a bicycle and rider is the discontinuation of the body. There are large areas with no solid body (eg wheel rim to hub, frame tube triangles, clearances between tyres and frames). This leads to inevitable separation of the freestream from the body surface and results in aerodynamic buffeting or aeroelastic effects (flutter). This causes the flow to take a long time to settle out and inevitably in that time, another variable has changed and the process repeats.

To illustrate the impact of transient drag, the graph below shows the yaw angle which is incremented in step inputs by 2 degrees every 10 seconds (data labels shown). This is plotted against drag force in steady state and in transient states.

The steady state line shows the drag performance of the wheelset when readings are allowed to settle and then noted.

The transient lines are more representative of real life. In the case of this data acquisition, a datum yaw angle was established and 2.5 deg/s of movement was overlapped. As the oscillation was introduced, there was an immediate increase in drag on both sets of wheels. At 4 degrees of yaw, there was a noticeable difference between the Reynolds and FLO wheels. The Reynolds wheels were able to deal with the instability and buffeting much better than the FLO wheels. Beyond 12 degrees, neither wheel was able to effectively contain the buffeting and full separation occurs.

In almost every case, the drag in the real world is much more than in a steady state scenario. It is particularly prevalent on the wheels because they are rotating and the net velocity at the top of the wheels are double the forward velocity.

Transient vs Steady State Drag

Time spent at varying Yaw Angles

Whilst the primary aim of this study was to establish a wind tunnel protocol to depict road analysis. Some of the data gathered could be used for generic calculations.

The instrumentation used for the road analysis had a sampling rate of 1024 times per second. Combining this level of accuracy with standard filtering protocols, it was possible to ascertain the effective yaw angle of the bike and rider. By reducing the resolution, the data was converted into a format that aligns with wheel manufacturer marketing departments for yaw angle vs time at that angle. In doing so, it reduced the accuracy of the results but has been shown for comparison purposes.

It should be noted that the transient data was a better reflection of actual time at an angle as it took into account the micro corrections for rider input steering and the instantaneous corrections for wind speed. Filtering for steady state by reducing the sample rate removed the instability. In summary the drag response against rate of change of yaw angle is a better predictor of response in a free stream at angles below the separation point of the section.

When considering an entire bike and rider combination, the effect of the wheels are comparatively small compared to the drag caused by the rider so the transient nature of the wheel drag becomes diminished. The rider drag is by far the dominant part of the system. The effects of transient response diminish as the ratio of forward to swirl (crosswind) velocity becomes greater. Thus the faster the rider goes, the less effect the transient behaviour becomes.

Time at Yaw Angle 30km/h
Time at Yaw Angle 50km/h
Time at Yaw Angle Combined

The effect of Tyre width on Aerodynamic performance

There has been a general trend towards wider tyres in the bicycle industry over recent years. This has been largely pushed by tyre and wheel manufacturers heading towards tubeless designs on the premise that a wider tyre has lower rolling resistance. Whilst the effects of rolling resistance and a more favourable contact patch have been well documented, the effect on aerodynamic drag has been disputed. Some wheel manufacturers have claimed their wheels were more aerodynamic with wider tyres – for this claim to be valid the wheels would have required a lower combined drag coefficient to overcome the increase in frontal area.

The graphs below show the comparison between two wheels, a narrow bodied Shimano C60 and a wider bodied Enve 7.8. It was clearly evident that a ballooned tyre (25mm on a Shimano C60 rim) had a significant impact on drag, especially at higher speeds. In contrast the effect on the wider bodied Enve wheel was much less dramatic. In both cases a narrow tyre reduced the drag. The continental tyres tended to measure slightly wider than their stated width when mounted.

Tyre width Drag 30km/h
Tyre width Drag 50km/h

Interpretation of the data

This data should be interpreted like those of fuel consumption figures for a car. They are designed to give a typical indication of how much power is absorbed over an ENTIRE ride loop at a given speed. It is important to note that wheels that are fast at 50km/h aren’t necessarily the fastest at 30km/h.

  • The MAXIMUM EXPERIMENTAL ERROR has been calculated at +/- 2.5%, the middle of the range is plotted for each of the values to maintain consistency
  • The rim depths are split into classes to make it easier for comparison, they may not agree with the stated size from the supplier.
  • The Power rating in a transient analysis is much worse than a steady state analysis
  • Comments are indicated for anything noteworthy
  • The rider position was within +/-10mm for each run, this was backed up with a reverse mounted pressure rake to remove any spurious data. There have been comments suggesting a rider could not keep a fixed position for the entire cycle duration – the protocol does not require them to, error checking is built in. Although unusual for the cycling industry, removal of appendage drag is commonplace in the Aerospace industry so the same technique was applied
  • Control Tyre was a pair of Continental GP4000SII 23mm with a pressure of 8.25BarG, there are two wheel combinations that differ from OE, these are a modified Borg wheel and a Mavic Tubular.
  • The ROTATIONAL Drag required to spin the wheel up is included (Most manufacturers do not include this figure which is around 25 to 30 percent, a notable exception is Swiss Side)
  • The riding position (relaxed hoods) remained unchanged irrespective of speed. In reality high speeds would necessitate a different riding position but doing so would have invalidated the test.

Who tests in Transient Conditions?

To date, the only company that has confirmed they test for and include turbulence (transient conditions) when designing is SwissSide. Jean-Paul Ballade of SwissSide commented on the Youtube video shown above.

Although unconfirmed, there are certain features on Mavic wheels which suggest they either design for or test in Turbulent conditions.


Riders have long been fed a diet of wheels being tested at 50km/h, this speed is inappropriate for the vast majority of riders as they cannot maintain the power required for that velocity. There has been a general thought process that most riding happens at yaw angles of less than 10 degrees. Whilst this may be a valid statement if you are doing 50km/h, at more modest speeds this does not occur. In both 50km/h and 30km/h riding, the effect of micro corrections to the steering, turbulence from the wind itself and external objects causes unsteady turbulent flow over the wheels. This phenomena causes the effective yaw angle experienced by the wheels to increase.

  • Wheels that performed well were noticeably resistant to generating areas of turbulence
  • Wheels that performed well mitigated generated turbulence quite well
  • Wheels that performed well had a lower rotational drag compared to their competition
  • Wheels with a deeper rim section are generally more aerodynamic than shallow sections
  • The difference between wheels of a similar depth is very small and it would be difficult for a human to be able to detect this during riding
  • The difference between a low profile wheel and a deep wheel would be picked up by a human riding.
  • The FLO cycling and Hunt wheels performed badly, they appear to have been designed by individuals with a limited understanding of aerodynamics of rotating objects. As such they generated unnecessary separation and could not deal with the separated airflow
  • The Aerocoach disc and 75mm deep section front wheel showed quite interesting results. This wheel was essentially an aluminium wheel with a clip on fairing. At low to moderate speeds, the wheel performed reasonbly but as the speed was increased the wheel started to perform quite erraticaly. The front wheel construction is agriculturual and large gaps exist between the spokes and the non structural fairing. These gaps generated pressure disturbances and caused the flow to behave erratically. As the speed was increased, it’s performance became quite poor in comparison to the immediate competition and this was mainly due to the poor front wheel design. A picture of the problem is shown below
Aerocoach Aeox Wheel

If you are considering using the data from this article to influence your purchasing decision then please use this with caution. Some aspects of wheels like their general build quality, braking performance, hubs and ease of maintenance are not measured. These factors should be taken into account accordingly.

Flo Cycling examined in more detail…

After publishing this data, some people commented the data was controversial. In the sense the brands they expected to be good weren’t. Of note was Flo Cycling. A brand that is run by two Canadian Entrepreneurs from their base in Las Vegas, they buy in rims from the far east and sell them with their markup to consumers around the world. Some further anlysis was conducted into the aerodynamics by Airbus aerodynamacists and the conclusion was unanimous.

It appears that Flo Cycling did not model or take into account the effect of the spokes on their wheels. The spokes and internal geometry contributes up to 40% of the total drag – the amount decreases as the wheels get deeper. Their wheels had very poor aerodynamic behaviour aft of the wheel rim.

FLO were asked if this accusation was true on several occasions, and on each occasion, they have declined to respond.

FLO’s marketing has a heavy engineering bias and to the untrained eye, this reads well. However when looking at their hub design, the calculations behind them and their engineering practices some gaping holes were noted.

When looking at this page Flo Cycling (Update: They deleted it so here’s a copy that was downloaded for evidence). It was quite obvious that the engineering behind flo wheels is commercially driven and not mechanically driven. Topics of note

  • The choice of measuring tool – a vernier caliper is totally unsuitable for measuring down to 0.01mm
  • The L10 calculation does not account for any axial loading in the bearing, to put this into layman’s terms, it does not account for the wheel going around corners or being subject to road vibration
  • If the bearing met ISO, JIS or DIN standards, it would not measure an exact mm reading, it would be slightly under (i.e. 20mm nominal is 19.99mm). This highlights the use of a poor measuring instrument or poor quality bearings
  • EZO bearings have only been chosen because the hub supplier in Taiwan uses them as an upgrade over the no name standard bearings they use. EZO bearings have been chosen for commercial reasons not technical reasons.

Here is what NTN technical had to say

The calculation provided from this page does not approximate the road behaviour of a wheel bearing with sufficient accuracy. One would expect to see the axial loads applied with a ramp or step input along with sinusoidal input for the radial loading. We would not expect the wheel bearing to last as long as the L10 life predicts because the boundary conditions for the data are inaccurate. Further, vernier calipers are not recommended by NTN for the measurement of our bearings, we cannot comment for other manufacturers.

FOLLOW UP… Letters from Lawyers

Shortly after writing this blog post, I received a letter from a firm of solicitors representing FLO cycling which was addressed to the HR department of my employer. They complained this page depicted their wheels in a poor light, the test protocol was not openly published and they did not like the statement that they had a limited understanding of the aerodynamics of rotating objects. They wanted their power figures removed from the data along with threats of court action. Furthermore, they requested I was dismissed from my Engineering position for misuse of company resources.

It should be noted that FLO Cycling have a somewhat questionable strategy of paying prominent forum members in a number of popular cycling/tri forums to endorse their products. They usually do this under the premise of supplying wheels.

FLO cycling’s behaviour has been disappointing as when the wheels were originally tested and an issue was found, FLO were contacted. This was 6 weeks prior to the data being uploaded. Once the data had been uploaded, they were first to comment and question every aspect of the testing in surgical detail, even going so far as to suggest an adjustment of 5% in tyre pressure would make their wheels more aerodynamic. As time has progressed, they have come out and openly stated that these results were fabricated.

Link to Letter from FLO Cycling

This Post Has 6 Comments

  1. Mike

    Very good read thank you!

    Wish you could have had some Campagnolo Wheels in there too…especially the new WTO series

  2. ka314159

    Great read, thanks, and I have a specific question if you have the time. When I was aero tested in a wind tunnel earlier in the year, they gave me a fairly consistent CdA figure of 0.25 riding on the drops for my track setup (Dolan Pre Cursa non-aero bike, skinsuit, regular helmet, deep section 80mm clinchers with 25mm Corsa G+). Recently when I was doing some solo efforts at the Manchester velodrome I found that my average power for two laps on the black line at 17s/lap (just under 53 kph) was 575w, which gives an implied CdA of 0.27 (Assuming Air density 1.226, Crr 0.004) . I just realized after reading your article that you are actually doing a 360-degree turn on each lap around a velodrome, so the yaw angle around the bends must be quite significant. Do you think the mismatch (0.25 vs 0.27 CdA) could be accounted for by the yaw, or could the differences in conditions and that my Crr is really just a guess, account for the differences?

    1. Francisco

      Hi ka314159, I think your question can be answered in the negative on the strength of very basic geometric reasoning. Your velodrome has a 250 m track, and while i could not find the precise radius of the curved ends it is unlikely to be less than 23 m. This gives you a curvature of 2.5 degrees per meter of track.
      a. Assuming your wheelbase is 1 meter and that you travelled along the track at such a glacially slow speed that there is no centripetal component at all, you steering input in the curved sections would also be 2.5 degrees. In reality you will be going quite fast, banked strongly in the curves, and your steering input will be almost imperceptible. The vector difference due to track curvature will still be there but, distributed between front and rear and further diminished by the tilt of the bike, I expect the actual impact of track curvature on angle of incidence to be about 1 degree or less, which will have a very small effect on air drag.
      b. Assuming you have 46cm of shoulder width, your right arm will slice the air about 2% faster than your left arm in the curved track sections. This corresponds to a 4% difference in drag between your left and right, but that is the extreme difference and for the whole body the differences will mostly cancel out. I certainly cannot imagine whole-body drag in the curved sections of the track being more than a fraction of a percentage point higher than in the straights.
      c. Therefore, the 8% difference between the 0.25 and 0.27 coefficients cannot be attributed to yaw or to the curvature of the track.
      Hope this helps. Better still, I hope our gentle host steps in and blows my analysis to bits.

Leave a Reply