A new maximum likelihood (NMLE) criterion suitable for model calibration using data which are recorded at unequal time intervals and contain autocorrelated errors is derived. Validation of the NMLE criterion has been carried out both on a simple two-parameter reservoir model using synthetically generated data and on a more complicated hillslope model using real data from the Pukeiti catchment in New Zealand. Comparison between the NMLE criterion and the simple least squares criterion reveals the superiority of the former over the latter. Comparison between the NMLE criterion and the maximum likelihood criterion for the autocorrelated case proposed by Sorooshian (1978) has shown that both criteria would yield results with no practical difference if equal time interval data were used. However, the NMLE criterion can work on variable time interval data, which provide more information than equal time interval data and therefore produce better visual results in hydrologic simulations.