Systematic variations of the IMF in early-type galaxies, and their connection with possible drivers such as velocity dispersion or metallicity, have been much debated in recent years. Strong lensing over galaxy scales combined with photometric and spectroscopic data provides a powerful method to constrain the stellar mass-to-light ratio and hence the functional form of the IMF. We combine photometric and spectroscopic constraints from the latest set of population synthesis models of Charlot & Bruzual, including a varying IMF, with a non-parametric analysis of the lens masses of 18 ETGs from the SLACS survey, with velocity dispersions in the range 200-300 km s −1 . We find that very bottom-heavy IMFs are excluded. However, the upper limit to the bimodal IMF slope (µ < ∼ 2.2, accounting for a dark matter fraction of 20-30%, where µ = 1.3 corresponds to a Kroupa-like IMF) is compatible at the 1 σ level with constraints imposed by gravity-sensitive line strengths. A two-segment power law parameterisation of the IMF (Salpeter-like for high masses) is more constrained (Γ < ∼ 1.5, where Γ is the power index at low masses) but requires a dark matter contribution of > ∼ 25% to reconcile the results with a Salpeter IMF. For a standard Milky Way-like IMF to be applicable, a significant dark matter contribution is required within 1R e . Our results reveal a large range of allowed IMF slopes, which, when interpreted as intrinsic scatter in the IMF properties of ETGs, could explain the recent results of Smith et al., who find Milky Way-like IMF normalisations in a few massive lensing ETGs.