Techniques for learned user preference- and behavior-based notification filtering are described herein. In one or more implementations, notifications obtained from computer applications are filtered for presentation to a user. Example notifications include notifications about emails, text messages, phone calls, web-page specific messages, antivirus application messages, and so forth. As part of filtering the notifications, interactions of a user with the notifications and with events for which the notifications can be generated are monitored. The monitored interactions are used to learn user preferences and behaviors for notifications in different contexts of user interaction with computing devices. Data is collected that describes characteristics of a current context. Based on the current context, importance scores are computed for new notifications using the learned user preferences and behaviors. The importance scores can then be used to determine which of the new notifications to present to the user.