The study tried to find out the best ARIMA model that could be used to make efficient forecast of inland, marine and total fish production in Bangladesh. It also tried to find out the best deterministic type growth model that are commonly used to describe growth pattern and also for forecasting. It appeared from the study that all the time series were second order homogeneous nonstationary. The total and inland production followed a second order integrated moving average process of order one. The process of generating the marine production was different from the other two processes. The second differenced marine production was simply a white noise. The fitness of the models was very satisfactory and forecasts obtained using them had very low standard errors. The forecasted total fish productions in 2000-01 to 2000-05 are 1763, 1867, 1974, 2085 and 2199 thousandlons. The best deterministic type growth models were quadratic for total and inland fish production and cubic for marine production. It also appeared that ARIMA models are the best for short term forecasting compared to the deterministic models.