The problem of scheduling parallel batch processing machines is considered from a clustering per spective in this paper. We first demonstrate that the batching problem with non identical job sizes can be re garded as a generalized clustering problem , providing a novel insight into scheduling with batching. The con cept of WR (waste ratio of batch space )is then presented and the objective function of minimizing makespan is transformed into minimizing weighted WR so as to define the distance measure between batches in a more understandable way. The equivalence of the two objective functions is also proved. In addition , a clustering algorithm CACB (constrained agglomerative clustering of batches )is proposed based on the definition of WR to generate batches. The experimental results show that CACB outperforms the existing approaches BFLPT (best Fit longest processing time )and GA (genetic algorithm )in large scale problems.