This thesis presents an online handwritten character recognition system for Urdu handwriting. The main target is to recognize handwritten script inputted on the touch screen of a mobile device in particular, and other touch input devices in general. Urdu alphabets are difficult to recognize because of inherent complexities of the script. In a script, Urdu alphabets appear in full as well as in half-forms: initials, medials, and terminals. Ligatures are formed by combining two or more half-form characters. The character-set in half-forms has 108 elements. The whole character-set of 108 elements is too difficult to be classified accurately by a single classifier. In this work, a framework for development of online Urdu handwriting recognition system for smartphones has been presented. A pre-classifier is de signed to segregate the large Urdu character-set into 28 smaller subsets, based on the number of strokes in a character and the position and shape of the diacrtics. This pre-classification allows to cope with the demand of robust and accurate recognition on processors having relatively low computational power and limited memory available to mobile devices, through banks of computationally less com plex classifiers. Based on the decision of the pre-classifier, the appropriate classi fier from the bank of classifiers is loaded to the memory to achieve the recognition task. A comparison of different classifier-feature combinations is presented in this study to exhibit the features’ discrimination capability and classifiers’ recognition ability. The subsets are recognized with different machine learning algorithms such as artificial neural networks, support vector machines, deep belief networks, long short-term memory recurrent neural networks, autoencoders-support vector machines, and autoencoders-deep belief networks. These classifiers are trained with wavelet transform features, structural features, and with sensory input val ues. Maximum overall classification accuracy of 97.2% has been achieved. A large database of handwritten Urdu characters is developed and employed in this study. This database contains 10800 samples of the 108 Urdu half-form characters (100 samples of each character) acquired from 100 writers.