Solar cells based on polymer-fullerene blends are amongst the most efficient organic photovoltaic devices with power conversion efficiencies now exceeding 3%. The large interfacial area in such dispersed heterojunctions, which is essential for charge separation, also enables charge recombination. Understanding recombination is, therefore, of key importance in the optimisation of device performance. Recent measurements of charge recombination in blends of poly(2-methoxy-5-(3',7'dimethyloctyloxy)-1-4-phenylene vinylene), (MDMO-PPV) and 1-(3-methoxycarbonyl)-propyl-i-phenyl-(6,6)C-61 (PCBM) by transient optical spectroscopy reveal that recombination dynamics possess two phases, one fast, intensity dependent phase and a slow, intensity independent phase which dominates at low light intensities. The recombination is thermally activated, and the kinetics are sensitive to background illumination but insensitive to the blend composition. Simple models of bimolecular recombination cannot explain these features. In this article, we present a model for the mechanism of charge recombination, based on multiple trapping of polarons by a distribution of traps in the polymer phase. The model explains the observed recombination kinetics and their dependence on light intensity, temperature and fullerene concentration. We show that under solar illumination conditions, charge recombination is limited by the activation of positive polarons out of deep traps, yet carrier collection competes successfully with recombination in thin films. The model is evaluated by comparison with data on cells and independent measurements of charge carrier mobility, and the implications for cell performance are discussed. (C) 2003 Elsevier B.V. All rights reserved.