A technique for the classification and understanding of child-like sketches of animals, using a live pen-based input device is demonstrated. A method of segmenting the sketch using curve differentials in addition to pen speed is shown. Once this stage is completed, an agent-based search of the resultant data begins in an attempt to isolate recognisable high-level features within the sketch. The context between probable matches for these features is used by the agents in assessing their success in identifying features correctly.