Timing control mechanisms were investigated in a complex rhythmic tapping task. Subjects performed a syncopated polyrhythm that consisted of two conflicting pulse trains (e. g., three elements per pattern repetition versus four elements per pattern repetition). The correlations between adjacent interval durations were found to differ systematically with the tempo at which the polyrhythm was performed. Only at fast tempos were subjects able to simultaneously control two different pulse trains. Alternative models are proposed that differ in their assumptions about the underlying pulse train. A comparison of the models suggests that the underlying pulse train is the faster one.