The present study evaluates the forecast accuracy, validity of rational expectation hypothesis and decomposition of forecast errors of the budget/revised budget estimates of federal and provincial governments of Pakistan for the sample period 1987-88 to 2011-12. The data has been taken mainly from various issues of the Annual Budget Statement (ABS). The data is analyzed through techniques of percentage forecast error, Theil inequality coefficient, regression analysis and mean squared prediction error. The mean squared prediction error has been decomposed into systematic and stochastic errors to trace the sources of errors. The results reveal that the forecast efficiency of the federal and provincial governments has not been satisfactory over the sample period; however it has been improved for some of the heads of revenue and expenditures. The decomposition of errors reveals the fact that stochastic variations contribute hugely to forecast errors, leaving a narrow room for improvement and the rational expectation hypothesis also does not hold for federal and provincial government budget forecasting. In the light of analyzed results, the study recommends the better forecasts of key macroeconomic variables, estimates of key parameters such as tax elasticity and stringent fiscal rules to combat forecast errors and enhance forecast efficiency. Supervisor:- Dr. Anwar Hussain