Handwritten Character Recognition is a long-standing problem among computer scientists. It promises several benefits to the user friendliness of modern computers. Several classical pattern recognition methods such as template matching, Fourier transformation, geometric moments and scene analysis which have proven to be effective in several domains have not yielded consistent or reliable results for handwritten character recognition Currently, neural networks are considered as the underlying computing mechanism for a robust approach to the problem of handwritten character recognition. This thesis presents various perspectives on pattern recognition techniques for handwritten character recognition using many different forms of neural networks. Six different hypotheses were investigated It is well known that many neural methods are overly specific to the training prototypes, a characteristic which is not suitable for the handwritten character recognition. A new neural network, called the polynomial network designed for more generalized classification is introduced. Apart from being more suitable this network performs classification of handwritten characters faster than standard backpropagation methods. Finally, this work presents a comprehensive collection of experimental evidence supporting claims for high character recognition abilities of Kohonen's learning vector quantization network, recurrent neural network, and the newly proposed polynomial neural network A body of theoretical and empirical results gathered during this work provides an insight to methodologies that may be the basis of future handwritten character recognition systems. In a broader sense the significance of this research extends to the general area of pattern recognition