Today's explosive growth of information demands personalized filtering. Collaborative filtering is a technique which provides personalized predictions on the ``likability'' of information based on the opinions of users who think alike. We suggest improvements to current collaborative filtering algorithms to increase the accuracy of the predictions. We implement correlation and history thresholds and conduct experiments that show our improvements increase the average prediction accuracy for all users, but require a per user threshold for optimum performance. These improvements can be used to either provide users with more accurate predictions or to provide them with an indication of the confidence they can place in a prediction.