In pervasive computing environments, devices often communicate in a peer-to-peer manner, either for passing messages or collaboratively running applications. For any particular device, information about the availability of other devices can lead to more efficient communication and better execution of distributed tasks. For example, a device can distribute computation tasks to reliable devices and temporarily avoid unreliable ones. Since there usually is not a central server to coordinate or monitor the communication in such environment, a key technical challenge is for each individual device to predict the availability of other devices. In this paper, we describe some methods for predicting the availability of other devices using historic availability data of those devices obtained in routine usage. In this scheme, each device separately maintains data about the past communication with the other devices, and predicts current and future availability using the statistics of those data. These methods do not require any centralized monitoring or extra probing, and thus have very low computation cost. These characteristics make them suitable for small devices in peer-to-peer environment.