A microlens of an inorganic material having a relatively high index of refraction is formed with a convex lower surface for refracting light from above through an underlying spacer layer to converge on a photodiode therebelow. The microlens and photodiode may be replicated in an array of such elements along with color filters and CMOS circuit elements on a semiconductor chip to provide an image sensor. The spacer layer, which has a relatively low refractive index, is subjected to a selective isotropic etch through an opening in an etch mask to define a concave surface that forms an interface with the convex lower surface of the microlens upon subsequent conformal deposition of the material of the microlens.