The behavior of smartphone apps is driven by input from sensors such as GPS, microphone, or camera. Hence the ability to construct test inputs, and send these inputs to the app is essential for testing. Leveraging our prior results in recording and replaying sensor inputs in Android apps we have constructed a new approach that helps automate smartphone app testing by capturing the input log (sensor stream) and using this log in two ways. First, we fuzz (alter) the log in a semantically-meaningful way: by applying principled transformations (e.g., changing GPS coordinates or navigation speed), a new input log is constructed, which represents a new test case. Second, we use the log captured in app A to test an app B which offers similar functionality, e.g., GPS navigation or image recognition. We have applied our approach to several widely-used Android apps and found that the approach is effective: it has revealed new bugs in four popular apps; has produced new test cases that increase coverage; and has produced test cases from logs originating in other apps.