Financial distress is an active research area particularly for business community of Pakistan due to economic conditions, electricity shortage and political situation. Banks are also taking keen interest in this area after the global financial crisis of year 2008. Therefore, the question that how financial distress can be predicted accurately has been widely debated by many scholars by using traditional statistical models. However, earlier research has not adequately addressed the issue of predicting financial distress. Adding to that the rate of financial distress is also getting harder to estimate by using traditional statistical models, because firms are becoming more complex and creating refined plans to hide their real financial situation. To prevent this condition latest prediction models are adopted by many countries which can give early indication of firm?s financial distress with highly accurate results. In this regard, prediction of financial distress by Neural Network Model is not much explored in Pakistan for foreseeing the financial health of firms. This paper addresses this issue and uses Neural Network Model to predict financial distress of firms in Pakistan by selecting suitable independent variables. The sample of 22 private sector conventional banks listed at Pakistan Stock Exchange is selected. The time series financial statements of these banks are selected for 15 years (2001 to 2015).Selected sample time frame is (pre-crisis 2001-2007), (crisis 2008) and (post-crisis 2009-2015). To test first hypothesis,4 Altman''s ratios from revised Altman''s Z-Score Model are calculated from these financial statements of selected banks. This study used three layered Neural Network Model consisting of input layer, hidden layer and output layer. The 4 independent explanatory variables/ input are 4 Altman''s ratios and 1 dependent variable/output is probable financial distress. After determining the Neural Network architecture, cross-validation re-sampling procedure is used to train, validate, and test a Neural Network by using commerciallyavailable MATLAB software. The best and most appropriate Neural Networks model, constructed by combining input variables of 4 Altman''s ratios, resulted in the R value of 0.99 that shows a relatively high accuracy given the error ratio in the input variables. These results confirmed the second hypothesis. By testing third hypothesis, distressed and non distressed banks are correctly classified with reference to Altman?s ratio