Predicting the merger time-scale (τmerge) of merging dark matter haloes, based on their orbital parameters and the structural properties of their hosts, is a fundamental problem in gravitational dynamics that has important consequences for our understanding of cosmological structure formation and galaxy formation. Previous models predicting τmerge have shown varying degrees of success when compared to the results of cosmological N-body simulations. We build on this previous work and propose a new model for τmerge that draws on insights derived from these simulations. We find that published predictions can provide reasonable estimates for τmerge based on orbital properties at infall, but tend to underpredict τmerge inside the host virial radius (R200) because tidal stripping is neglected, and overpredict it outside R200 because the host mass is underestimated. Furthermore, we find that models that account for orbital angular momentum via the circular radius Rcirc underpredict (overpredict) τmerge for bound (unbound) systems. By fitting for the dependence of τmerge on various orbital and host halo properties, we derive an improved model for τmerge that can be applied to a merging halo at any point in its orbit. Finally, we discuss briefly the implications of our new model for τmerge for semi-analytical galaxy formation modelling.