The low-mass end of the stellar initial mass function (IMF) is constrained by focusing on the baryon-dominated central regions of strong lensing galaxies. We study in this Letter the Einstein Cross (Q2237+0305), a z = 0.04 barred galaxy whose bulge acts as lens on a background quasar. The positions of the four quasar images constrain the surface mass density on the lens plane, whereas the surface brightness (H-band HST/NICMOS imaging) along with deep spectroscopy of the lens (VLT/FORS1) allows us to constrain the stellar mass content, for a range of IMFs. We find that a classical single power law (Salpeter IMF) predicts more stellar mass than the observed lensing estimates. This result is confirmed at the 99 per cent confidence level, and is robust to systematic effects due to the choice of population synthesis models, the presence of dust or the complex disc/bulge population mix. Our non-parametric methodology is more robust than kinematic estimates, as we do not need to make any assumptions about the dynamical state of the galaxy or its decomposition into bulge and disc. Over a range of low-mass power-law slopes (with Salpeter being Gamma = +1.35) we find that at a 90 per cent confidence level, slopes Gamma > 0 are ruled out.