A novel analysis approach has been developed to examine the latency of distortion product otoacoustic emissions (DPOAEs). DPOAEs were measured in ten normal‐hearing adults in a paradigm in whichf2was held constant andf1was varied. This paradigm was used with a wide range of primary levels. Latency was estimated in two ways. In the first, a phase‐slope delay measurement was used which showed a significant response latency increase as stimulus intensity was decreased. In the second approach, an inverse‐FFT procedure was used to provide a temporal analysis of the data. Results of this analysis reveal a complex latency structure with multiple peaks in the envelope of the time waveform. The latencies of individual peaks remain constant across level, however short latency peaks have the greatest amplitudes at higher levels, and longer latency peaks are largest at low levels. These results would be consistent with the idea that there are multiple intracochlear sources for distortion product generation; however, a simple model, in which generation is assigned to thef2and the 2 f1−f2place, does not adequately explain the number of envelope peaks that were present in many ears.