We introduce Early Vision Networks (EVNets), hybrid CNNs combining biologically inspired subcortical and cortical front-end blocks, which improve V1 alignment and significantly boost robustness to different image perturbations.