In this paper, we propose a technique for generating particles from user-specified transfer function for an effective point-based volume rendering. In general, a volume rendering technique utilizes an illumination model in which the 3D scalar field is characterized as a varying density emitter with a single level of scattering. This model is related to a particle system in which the particles are sufficiently small and of low albedo. A conventional volume rendering technique models the density of particles, not the particles themselves [1]. The density is defined by specifying a transfer function from a scalar data value to an opacity data value. Thus, a given scalar field is described as a continuous semitransparent gel and the accumulating order is important. This results in a considerable computational overhead. On the other hand, our rendering technique represents the 3D scalar fields as a set of particles. The particle density is derived from a userspecified transfer function, and describes the probability that a particle is present at the point. Since the particles can be considered as fully opaque, no alpha blending but only depth comparison is required during the rendering calculation, which is advantageous in the distributed processing.