Water-bottom related multiple energy can be a significant problem for processing and interpretation of seismic data. The standard method of NMO stacking CMP gathers to eliminate multiples is often inadequate and does not address the problem of preserving interesting features on prestack data such as AVO and non-hyperbolic moveout. In this paper we examine primary and multiple energy in beam stacked data and a method of isolating the primary energy in this space. We decompose the data into the beam stacked space by means of an iterative least squares inversion of the beam stack operator. We then apply a masking function to the beam stacked data. The primary energy is then forward-modeled simply using the beam stacking operator. We apply our method to real data and obtain encouraging results.