Ren­der­ing a com­plex spec­u­lar sur­face un­der sharp point light­ing is far from easy. Us­ing Monte Carlo point sampling for this pur­pose is im­prac­tic­al: the en­ergy is con­cen­trated in tiny high­lights that take up a minus­cule frac­tion of the pixel. We in­stead com­pute the ac­cur­ate solu­tion that Monte Carlo would even­tu­ally con­verge to, us­ing a com­pletely dif­fer­ent de­term­in­ist­ic ap­proach with min­im­al ap­prox­im­a­tions. Our meth­od con­siders the true dis­tri­bu­tion of nor­mals on a sur­face patch seen through a single pixel, which can be highly com­plic­ated. This re­quires com­put­ing the prob­ab­il­ity dens­ity of the giv­en nor­mal com­ing from any­where on the patch. We show how to eval­u­ate this ef­fi­ciently, as­sum­ing a Gaus­si­an sur­face patch and Gaus­si­an in­trins­ic rough­ness. We also take ad­vant­age of hier­arch­ic­al prun­ing of po­s­i­tion-nor­mal space to quickly find texels that might con­trib­ute to a giv­en nor­mal dis­tri­bu­tion eval­u­ation. Our res­ults show com­plic­ated, tem­por­ally vary­ing glints from ma­ter­i­als such as bumpy plastics, brushed and scratched metals, metal­lic paint and ocean waves.