Catch data for the period 1950-91  on all India landings for three important marine fish species and total landings were analysed to develop suitable ARIMA models for time series forecast. An initial recursive regression analysis showed that the models suitable are ARIMA (2,2,1) for total landings,  ARIMA (1,2,1) for penaeid prawn landings and ARIMA (3,2,1) for both catfish and silverbellies landings. The parameters of these models were then estimated by a maximum likelihood estimation procedure following the algorithm of Godolphin (1984). A test by using Chi-square given by Lung and box (1978) showed that the fitted models are adequate to explain the data. With the fitted models forecasts for the next year were computed along with their confidence limits.