We propose a generic, hybrid constraint handling scheme for particle swarm optimization called Heterogeneous Constraint Handling. Inspired by the notion of social roles, we assign different constraint handling methods to the particles, one for each social role. In this paper, we investigate two social roles for particles, `self' and `neighbor'. Due to the usual particle dynamics, a powerful mixture of the two corresponding constraint handling methods emerges. We evaluate this heterogeneous constraint handling approach with respect to the complete set of the CEC 2006 benchmark instances. Our results indicate that a such a heterogeneous combination of two constraint handling methods often leads to significantly better results than running each individual constraint handling method separately and returning the best solution obtained.