This work introduces an image processing methodology that, associated with a multi-level neural network of perceptrons, is able to isolate and recognize cursive handwritten characters. The character isolation technique makes use of fundamental geometric and topological aspectos of the characters. The work describe procedures to extract the characters skeletons, such as thinning and smoothing heuristic algorithms, zoned filtering to attenuate horizontal and vertical lines, contour detection, heuristic extraction of characteristics and the computation of Fourier Descriptors representing the line patterns, that compose the characters. After character extraction, its combined characteristics are presented to a neural network in order to allow recognition (identification). Finally, the results of the character identification are combined to avoid classification intersections, due to common aspects in a number of characters. The introduced methodology concerns only with the segmentation and form identification of the characters. It does not adress any context analysis.