This document describes a study that uses machine learning algorithms to recognize similar shaped handwritten Hindi characters. It extracts 85 features from each character image based on geometry. Four machine learning algorithms (Bayesian Network, RBFN, MLP, C4.5 Decision Tree) are trained on datasets containing samples of a target character pair and evaluated based on precision, misclassification rate, and model build time. Feature selection techniques are also used to reduce the feature set dimensionality before classification. Experimental results show that different algorithms perform best depending on the feature set and number of samples used for training.