Over the past few years, Augmented Reality has become widely popular in the form of smart phone applications, however most smart phone-based AR applications are limited in user interaction and do not support gesture-based direct manipulation of the augmented scene. In this paper, we introduce a new AR interaction methodology, employing users' hands and fingers to interact with the virtual (and possibly physical) objects that appear on the mobile phone screen. The goal of this project was to support different types of interaction (selection, transformation, and fine-grain control of an input value) while keeping the methodology for hand detection as simple as possible to maintain good performance on smart phones. We evaluated our methods in user studies, collecting task performance data and user impressions about this direct way of interacting with augmented scenes through mobile phones.