Wireless sensors on rotating mechanical structures have rich and fast changing multipath that cannot be easily predicted by conventional regression approaches in time for effective transmission coding or power control, resulting in deteriorated transmission quality. This study aims to study the speed-dependent packet error rate (PER) of wireless sensor radios on rotating mechanical structures. A series of rotating IEEE 802.15.4 sensor radio transmission experiments and vector network analyzer measurements have been conducted to derive and validate a predictive PER model for a fast rotating sensor radio channel based on channel impulse response measurements. The proposed predictive PER model, including power attenuation, bit error rate (BER) and PER sub-models, captures the channel property of rotating sensors based on the received signal strength and the radio receiving sensitivity. The PER model has accurately predicted the PER profile of sensors on a rotating machine tool spindle as well as a rotating plate of a prototype rotation system. The analysis provides an in-depth understanding of how multipath propagation causes the fast power variation and the resulting speed-dependent PER for wireless sensors on rotating mechanical structures.