In today's multi-device computing environment a lot of different applications use notifications to inform the user about certain events such as new e-mails, available updates and more. This makes notifications an integral part of interaction with smart devices. However, related work showed that notifications have disruptive effects and are responsible for a large number of interruptions. Thus, with an increasing number of smart devices, including smartphones, tablets, smartwatches and desktop PCs, the negative effects of notifications are exacerbated in the omnipresent multi-device environments of today. As a consequence, there is the need to find a balance between keeping the user informed and confining the negative effects of notifications. For that reason, we developed a framework which is able to select a device or subset of devices to notify the user. Hence, we developed applications which collect contextual information and sensor data from the devices of the user. Furthermore, we conducted an in situ study using the Experience Sampling Method to gain insight into how devices in a multi-device environment are used and to find out which of the contextual information can be used to predict on which device or subset of devices the user would like to receive a notification. Our findings show that there is a strong correlation between the proximity of a device to a participant and whether a notification should be shown on that device. Furthermore, we found out that there are correlations between the collected information and the preferred displaying device for a notification. Based on our findings we implemented a server application for selecting the device or subset of devices as a proof of concept. As a result, the disruptive and interruptive effects caused by notifications in a multi-device computing environment can be minimized.