Summary: Some properties of a class of quasi-differentiable functions (the difference of two finite convex functions) are considered in this paper. The convergence of the steepest descent algorithm for unconstrained and constrained quasi-differentiable programming is proved.