In this paper the performance of predetection maximal ratio and equal gain combiners are investigated under conditions of correlated branch noise. A statistical model is devised to determine the spatial noise correlation coefficients at metropolitan-area base stations, and the cases where significant correlation is likely are clarified. Optimal weighting coefficients for a maximal ratio combiner with two-branch space diversity are derived under correlated noise. Based on this result it is shown that correlation in branch noise can be used to improve the combiner performance by dynamically adjusting the weightings so as to partially cancel the noise. Performance of equal gain combiners is also shortly discussed.