Automatic license plate recognition is useful in many contexts such as parking control, law enforcement and vehicle background checking. The high cost and low portability of commercial systems makes them inaccessible to the majority of end users. However, current mobile devices now have processors and cameras that make image processing and recognition applications feasible on them. This thesis investigates high accuracy real-time license plate recognition on a smartphone, taking into account device limitations. It first explores how, using the minimal image processing and simple configurable heuristics based on plate geometry, license plates and their characters can be detected in an image. Then, using minimal training data, it shows that a character recognition package can achieve high levels of accuracy. This approach accurately recognized 99 percent of plates appearing in a test set of videos of vehicles with New Zealand license plates.