The cerebellum is critical for locomotor coordination, but it is still not well understood how locomotor signals are represented within the cerebellum to enable whole-body coordination. In particular, while cerebellar Purkinje cell activity is strongly modulated by the locomotor stride cycle, it is not clear how that activity is constructed from its afferent inputs, nor how it can be used for real-time control. To investigate these questions, we are recording cerebellar population activity with Neuropixels probes during locomotion in head-fixed mice, combined with high-speed 3D tracking of limb and body movements. We used spike waveforms, spiking statistics, and probe location and depth to identify recordings from putative Purkinje cells, as well as their inputs and outputs. Consistent with previous work, we find that most units show modulation to the locomotor stride cycle. Moreover, different putative cell types show differential responses to locomotor events. These responses hint at distinct functional roles throughout the cerebellar circuit. Finally, we are investigating how cerebellar activity could be read out by downstream areas. Our preliminary analyses show that locomotor kinematics lie on a low dimensional manifold, whereas cerebellar activity appears to be high-dimensional and highly variable between strides. We find that we can accurately decode paw positions from instantaneous firing rates, showing that information about single trials is contained in the neural data. Together, our results suggest that locomotor coordination is readily decoded from cerebellar population activity, and that the high-dimensionality of this activity may help ensure robust control of motor behavior in variable contexts.