We address the problem of selecting sensors, that is, output equations, in order to endow dynamic equations of a system with some properties. Among all such desirable properties is the basic one of observability and/or identifiability. Once such a problem is solved one may ask how to choose sensors in order to improve estimation algorithms in terms of reliability, robustness, or simply, low complexity. First, what is the minimal number of sensors that make the dynamics observable? Second, when the sensors are bound to measure state components, what is their minimum number? Third, how may the observability margin be improved by selecting the sensors? In this communication we provide an overview of these questions in the differential algebraic approach of observation problems.