The development of G protein-biased agonists for the μ-opioid receptor (MOR) offers a clear drug discovery rationale for improved analgesia and reduced side-effects of opiate pharmacotherapy. However, our understanding of the molecular mechanisms governing ligand bias is limited, which hinders our ability to rationally design biased compounds. We have investigated the role of MOR binding site residues W320 and Y328 in controlling bias, by receptor mutagenesis. The pharmacology of a panel of ligands in a cAMP and a β-arrestin2 assay were compared between the wildtype and mutated receptors, with bias factors calculated by operational analysis using ΔΔlog(τ/KA) values. [3H]diprenorphine competition binding was used to estimate affinity changes. Introducing the mutations W320A and Y328F caused changes in pathway bias, with different patterns of change between ligands. For example, DAMGO increased relative β-arrestin2 activity at the W320A mutant, whilst its β-arrestin2 response was completely lost at Y328F. In contrast, endomorphin-1 gained activity with Y328F but lost activity at W320A, in both pathways. For endomorphin-2 there was a directional shift from cAMP bias at the wildtype towards more β-arrestin2 bias at W320A. We also observe clear uncoupling between mutation-driven changes in function and binding affinity. These findings suggest that the mutations influenced the balance of pathway activation in a ligand-specific manner, thus identifying residues in the MOR binding pocket that govern ligand bias. This increases our understanding of how ligand/receptor binding interactions can be translated into agonist-specific pathway activation.