The present experiment measured an EEG indicator of motor cortex activation, the lateralized readiness potential (LRP), while participants performed a speeded category classification task. The LRP data showed that visually masked words triggered covert motor activations. These prime-induced motor activations preceded motor activations by subsequent (to-be-classified) visible target words. Multilevel statistical analyses of trial-level effects, applied here for the first time with electrophysiological data, revealed that accuracy and latency of classifying target words was affected by both (a) covert motor activations caused by visually masked primes and (b) spontaneous fluctuations in covert motor activations. Spontaneous covert motor fluctuations were unobserved with standard subject-level (multi-trial) analyses of grand-averaged LRPs, highlighting the utility of multilevel modeling of trial-level effects.