Based on the mechanism of action, a quantitative structure-activity relationship (QSAR) model for the depuration rate constants (k(d)) of 28 PAHs, 8 PBDEs and 28 PCBs in mussels (Elliptio complanata) was constructed by partial least squares (PLS) regression, following the guidelines for development and validation of QSAR models. For the training set of the QSAR model, r² = 0.953, the cross-validated regression coefficient (Q²(CUM)) was 0.947. The predicted log k(d) values for the validation set were consistent with the observed values, with a standard error (SE) of 0.160 log units and a squared correlation coefficient (Q²(EXT)) of 0.892. Comparatively, the developed QSAR model had good robustness, predictive ability and extended applicability domain. The electrophilicity index (ω), molecular polarizability (α), the averages of the negative potentials on the molecular surface (V(s⁻)) and the balance parameter of surface potential (τ) were the key parameters governing the log k(d) values in the QSAR model, which indicated that the log k(d) value was mainly related to the partition ability, electrostatic interactions, and van der Waals interactions of compounds.