Analysis of behavior using video is a promising approach for identifying risk markers for psychopathology that can be applied in a wide range of populations. As part of a study on the environmental factors that relate to obsessive-compulsive disorder (OCD) behaviors, videos were recorded of everyday tasks being performed by two groups of children: a control group and a group diagnosed with OCD. One of the activities involved handwashing, since handwashing compulsions are frequent amongst those who suffer from OCD. Being able to classify these handwashing videos as showing behaviors associated with OCD or not is a step towards helping to automate important aspects of this psychiatric study. This paper explores using various feature descriptors sampled from dense motion trajectories to determine which combination of features and encodings would be best for video classification. Dense motion trajectories are computed for the videos from the OCD study and the points in these trajectories are described using several methods, including histograms of oriented gradient, histograms of optical flow, and motion boundary histograms. Various encoding techniques for these descriptors are also explored, including bag of words, pyramid bag of words, and sparse coding. To determine which feature/encoding techniques would perform the best, several dimensionality reduction techniques are used and the methods are ranked based on separability in the low dimensional space. This separability is measured by classifying using a linear discriminant and also by using kNN.