Gesture Interaction may prove useful for certain interaction scenarios; however if high recognition accuracies are not achieved then it can severely limit the production rate. Currently I am working with early recognition of user-defined gestures with Hidden Markov Model techniques. Hidden Markov Models are a specific case of Weighted Finite State Acceptors that assume a set of state transitions responsible for an observable sequence of features of a gesture. Recognition rates are high for small gesture sets and low for large gesture sets. Due to the spatio-temporal variability of gestures, this creates feature space variations that normal Hidden Markov Models model as noise. Despite the variety of Hidden Markov Models, and other learning techniques the recognition rate is still too low for larger gesture sets. Early Recognition and higher recognition accuracies are the goals of this research project.