The back-propagation algorithm suffers with the local minima error surface for a large set of problems. A local minimum is defined as a point such that all points in a neighborhood have an error value greater than or equal to the error value in that point. These regions of local minima occur for combinations of the weights from the inputs to the hidden nodes such that one or both hidden nodes are saturated for at least two patterns. However, boundary points of these regions of local minima are saddle points. This paper describes the soft computing techniques for the performance evaluation of Back-propagation algorithm to recognize the hand written English alphabets. Two different architectures of neural network have been taken with five trials of each network. It has been analyzed that the conventional back-propagation algorithm suffers with the problem of non-convergence of the weights. The results of the experiments and result shows that the conventional back-propagation algorithm does not suites to solve the challenging problem most reliably and efficiently.