The problem of automatic recognition of scanned handwritten documents is of great importance in scientific, business, and personal applications. This work develops novel methods to generate synthetic handwriting by computer, which enable handwriting recognizers to be better trained with the help of synthetically expanded sets of training samples.