Neuronal activity of prefrontal cortex (PFC) was examined during a duration discrimination task using visual (Vis, a green square) and auditory (Aud, 2000Hz tone) stimuli in two monkeys. In the task, two cues (first cue, C1 and second cue, C2) were presented consecutively for different duration ranging from 0.2 to 1.8 sec. Each cue was followed by 1 sec delay period (first delay, D1 and second delay, D2). Subjects were required to choose the longer presented cue after the D2 period. Four kinds of cue modality combinations were adopted; Vis-Vis, Vis-Aud, Aud-Vis and Aud-Aud. Duration of two cues, order of cue duration (long-short, LS or short-long, SL), and cue modality combination were pseudo-randomized. Correct rates were significantly different among the four modality combinations in both monkeys. Those for the Aud-Aud trials were lower than other three types. In the Aud-Aud and Aud-Vis trials, correct rates were significantly lower during the LS trials than the SL trials. Out of 860 PFC neurons examined, 64, 35, 139 and 192 neurons were C1, D1, C2 and D2 responsive, respectively. More than 80% of C1 response neurons were modality-specific, responded to either the visual or auditory C1, while 14% of the neurons were bimodal. Among 35 D1 response neurons, 7 and 10 neurons changed their activity depending on durations of the visual and auditory C1, respectively. Out of 192 D2 response neurons, 57 neurons exhibited greater D2 activity in the LS trials, while 85 neurons did so in the SL trials. About 40% of D2 response neurons responded differently after the visual and auditory C2. These D2 response neurons represent relative order-based timing, whether C1 or C2 was presented longer, partially in modality-dependent manner. Recording area was mainly on the lateral side of the principal sulcus. These results suggest that PFC integrates duration information on different modes of sensory stimuli and contributes to temporal discrimination processing of these stimuli, partially in parallel neural networks for different modalities. This work was supported by grants from the Ministry of Education, Culture, Sports, Science and Technology of Japan (21500376, 24500331).