We present a new and simple data-driven technique for modeling 3D brushes for use in realistic painting programs. Our technique simplifies and accelerates simulation of the constrained dynamics of brushes by using a small lookup table that efficiently encodes the range of feasible constrained states. The result is a brush model which runs an order of magnitude faster than previous physicsbased methods, while at the same time delivering greater deformation fidelity.