Character recognition plays an important role in the modern world. Character recognition is considered to be very important solution to many problems because of its numerous applications and theoretical values in the domain of pattern recognition. The only real problem is to design a system that can extract the right features efficiently. Thus, most of the attention is devoted to find the optimal sequence of preprocessing techniques, character segmentation procedure, and efficient feature extraction methods. In this thesis important operations are used in preprocessing of the input image in a way that increases the character recognition rates. A number of steps are used in the work namely: Histogram stretching, Mean Mask and finally Binarization technique. This is followed by segmentation process and fixing character boundaries. Number of feature extraction techniques has been used to obtain the features. These are Discreet Cosine Transform (DCT), and two Wavelet Transform (WT) techniques (Haar and Daubeche-4 or Db4).Pattern matching with database for letters and decimal digits using Euclidean Distance as decision metric. Recognition rates are computed and used as performance measure for each method. The results have shown that the use of DCT method gives the best results when 25 or more coefficients are involved in the feature extraction. Using larger number of coefficients require more processing time. On the other hand wavelet based techniques provide slightly less recognition rates. These techniques are tested with very small number of coefficients.