The presence of heavy metals and metalloids in water represents nowadays one of the most important environmental problems. These species have infinite lifetimes, and chemical or biological treatments present severe restrictions or are economically prohibitive. Hexavalent chromium, mercury, uranium, arsenic, or lead are on the list of priority pollutants of most environmental agencies, with more and more exigent limits of discharge or concentration in drinking water. From the beginning of the development of heterogeneous photocatalysis, the transformation and deposition of metals or metalloids were visualized as processes with promissory potential application to remove these pollutants from water. Three types of mechanisms can be considered for these processes, all of them taking place through successive monoelectronic electron transfer steps: (a) direct reduction by photogenerated electrons; (b) indirect reduction by intermediates generated from electron donors (reducing radicals); (c) oxidative removal by holes or hydroxyl radicals. This presentation is an overview of the work performed in our laboratories with the cases of hexavalent chromium and arsenic being treated in profundity. In the case of hexavalent chromium, direct reductive photocatalysis and indirect reduction by intermediates coming from ethylenediaminetetraacetic acid (EDTA) or citric acid added as electron donors are the main processes governing the removal of Cr(VI) by TiO2 photocatalysis. In the case of arsenic, removal can proceed by oxidation of As(III) to As(V), a very much studied process. However, reductive photocatalysis has been less studied and can take place under specific conditions, leading to the removal of As species by the formation of As(0) on the surface of the photocatalyst. While for As(III) direct reduction by photogenerated electrons is possible, As(V) reduction only proceeds in the presence of an electron donor such as methanol. The mechanisms taking place in these cases will be postulated in this presentation and the possible application to real systems will be discussed.