A genetic programming approach to fuzzy logic controller (FLC) design is presented in this paper. We propose an encoding that represents fuzzy rule- bases as type-constrained syntactic trees implemented as variable-length strings. This encoding is applied to the cart centering problem and compared with other approaches: an intuitive FLC done by hand, an FLC obtained by a traditional genetic algorithm operating on fixed length strings and the analytical optimal solution (bang-bang rule). The results of the GP model are better (158 time steps) than the intuitive solution (249 time steps) and comparable with the GA solution (149 time steps). The analytical optimal solution docks the cart in 129 time steps.