Different models of the binaural system make different predictions for the just-detectable interaural time difference (ITD) for sine tones. To test these models, ITD thresholds were measured for human listeners focusing on high- and low-frequency regions. The measured thresholds exhibited a minimum between 700 and 1,000 Hz. As the frequency increased above 1,000 Hz, thresholds rose faster than exponentially. Although finite thresholds could be measured at 1,400 Hz, experiments did not converge at 1,450 Hz and higher. A centroid computation along the interaural delay axis, within the context of the Jeffress model, can successfully simulate the minimum and the high-frequency dependence. In the limit of medium-low frequencies (f), where f . ITD << 1, mathematical approximations predict low-­ frequency slopes for the centroid model and for a rate-code model. It was found that measured thresholds were approximately inversely proportional to the frequency (slope = –1) in agreement with a rate-code model. However, the centroid model is capable of a wide range of predictions (slopes from 0 to –2).