Evidence is provided from forward masking conditions of an auditory detection study demonstrating that an early temporal integration system summates auditory information for 3 msec or less. The output of this early system is the input to another system that applies a compressive nonlinear transformation and summates auditory information for a considerably longer period of time.