Abstract This paper investigates the efficacy and reliability of three different state-of-the-art rotation speed estimation techniques on a very large set of experimental vibration data originating from thirty offshore wind turbine gearboxes. The three methods include the multi-order probabilistic approach, the phase demodulation method based on the frequency-domain energy operator, and the multi-harmonic demodulation technique. The goal is twofold: to assess statistically the performance of present-day vibration-based rotation speed estimation techniques on challenging experimental data, and to establish indirect rotation speed estimation through vibration data as a viable alternate solution to the conventional solution involving a direct measurement with a physical device such as a tachometer or an angle encoder. The results show that while all three techniques attain satisfying results, the multi-harmonic demodulation technique produces the most accurate speed estimate for the majority of all measurements whilst also being flexible in its usage.