Watch-list-based classification and verification is advantageous in a variety of surveillance applications. In this thesis, we present an approach for verifying if a query image lies in a predefined set of target samples (the watch-list) or not. This approach is particularly useful at identifying a small set of target subjects and therefore can render high levels of accuracy. Further, this approach can also be extended to identify the query image exactly out of the target samples. The three- stages approach proposed here consists of using a combination of color and texture features to represent the image and further using, Kernel Partial Least Squares for dimensionality reduction followed by a classifier. This approach provides improved accuracy as shown by experiments on two datasets.