In this paper we present an approach to define reusable gestures for embodied web agents, given according to the H-anim standard for VRML. We identify the dimensions to compare gestures and criteria to circumscribe a set of gestures an avatar should be endowed with. Based on these dimensions and criteria, we propose a uniform way to define a wide range of gestures that may be adapted to particular circumstances and personal style. We have implemented our gesture library in STEP, which is a scripting language for humanoids based on dynamic logic. We will discuss the example of conducting music, to explain our approach to the parameterized definition and control of gestures. We conclude by briefly discussing issues for future research.