We investigate the effect of primordial non-Gaussianity of the local fNL type on the auto- and cross-power spectra of dark matter haloes using simulations of the Λ cold dark matter cosmology. We perform a series of large N-body simulations of both positive and negative fNL, spanning the range between 10 and 100. Theoretical models predict a scale-dependent bias correction Δb(k, fNL) that depends on the linear halo bias b(M) . We measure the power spectra for a range of halo mass and redshifts covering the relevant range of existing galaxy and quasar populations. We show that auto- and cross-correlation analyses of bias are consistent with each other. We find that for low wavenumbers with k < 0.03 h Mpc−1 the theory and the simulations agree well with each other for biased haloes with b(M) > 1.5 . We show that a scale-independent bias correction improves the comparison between theory and simulations on smaller scales, where the scale-dependent effect rapidly becomes negligible. The current limits on fNL from Slosar et al. come mostly from very large scales k < 0.01 h Mpc−1 and, therefore, remain valid. For the halo samples with b(M) < 1.5 − 2 , we find that the scale-dependent bias from non-Gaussianity actually exceeds the theoretical predictions. Our results are consistent with the bias correction scaling linearly with fNL.