The invention discloses a fingerprint direction field calculation method based on an improved partial differential model, and belongs to the technical field of automatic fingerprint recognition. The fingerprint direction field calculation method mainly comprises two steps: I, solving an initial direction field of fingerprint by using a basic gradient algorithm; II, smoothing the initial direction field by using the improved partial differential model. The fingerprint direction field calculation method mainly aims to improve holding items of a conventional partial differential model, that is, firstly, the situation that the initial direction field in the conventional model is taken as holding standard is improved into the situation that the direction field of previous iteration is taken as the holding standard, and secondly, a static capture singular point function in the conventional model is improved to be a dynamic capture singular point function. By adopting the fingerprint direction field calculation method disclosed by the invention, noise areas and singular point areas can be effectively distinguished, the algorithm is relatively reasonable, the iteration times can be effectively reduced, the algorithm efficiency can be improved, and the direction field can be relatively accurately calculated.