A method and system for training a handwriting recognizer at the time of misrecognition is provided. In a preferred embodiment, a trainer program traps and stores the output from a recognizer. The output includes the recognized symbols mapped to the handwritten data. When training is requested by a user, the trainer displays the handwritten data that was input by the user and the recognized symbols. The user then selects the symbols to train on. In response to the selection, the trainer displays the portion of the handwritten data corresponding to the selected symbols. The user then enters the symbols that the displayed portion of the handwritten data should be recognized as. The trainer then requests the recognizer to train based on the displayed portion of the data and the entered symbols. When training is complete, the trainer requests re-recognition of the handwritten data and displays the newly recognized symbols.