This article maintains that technical issues for sampling in epidemiological surveillance require a profound review of important concepts from health theory. It is particularly neccessary to emphasize living conditions and more specifically the environments or contexts in which reproductive processes of social life take place. However, both fields require access to more complete data than those produced by traditional sources. Such an approach to epidemiological surveillance requires a review of sampling types, and this implies a new look at prevailing interpretations concerning the logical basis for inferences from samples. It becomes necessary to abandon statistical samples (even stratified ones) and to promote procedures of the "sentinel site" type. This techinique, originally applied in societies with insufficient statistical systems, can be developed in such a way as to become a substantial complement to monitoring of living conditions, even in societies with good information systems. The article suggests changes in the "sentinel site" concept, adding the requirement of "qualitative representativeness" through finalistic samples based on previous typologies of spatial/demographic units.