This paper presents a quantitative functional model, called the correlation model, to be used for interpreting empirical results on binaural masking level differences (BMLDs), which are obtained experiments on the detection of sinusoidal signals embedded in binaural noise. The receiver is presumed to behave as if it computes a statistical decision variable equivalent to a linear combination of three quantities, the energy levels at the channels deriving from the two ears and the interchannel cross correlation, where the coefficients are dependent on the interaural noise cross correlation and the interaural amplitude ratio for noise but are completely independent of signal parameters. Additive internal noise is assumed. Equations for BMLDs are derived with the restriction of equal noise levels at the two ears. Predictions derived from the model are compared with empirical results from several studies. These show BMLDs for antiphasic, homophasic, and monoaural input configurations at different frequencies of tonal signal, and BMLDs for varying interaural correlations of signal and of noise, where the noise correlation is statistical or deterministic. Similarities and differences between the correlation model and another model, the equalization cancellation model, are discussed in detail.