Experimental or quasi-experimental studies of the behavior of staff working with retarded persons were reviewed and provided the following generalizations: (a) staff often have very low rates of interactions with clients, (b) a disproportionate number of interactions occur with a relatively small number of clients, (c) structured settings promote more interactions, (d) small teaching groups promote interactions, (e) staff involvement in relevant decisions promotes interactions, (f) larger, more global, and ecological assessments of both staff and clients provide important information on the effects of staff management and training, (g) lowering the reading difficulty of training materials increases staff comprehension, (h) lectures tend to improve academic skills while practicum training improves teaching skills, (i) some trained skills generalize while others do not, apparently unsystematically, (j) a system in which supervisors rather than researchers external to the facility do the training is feasible, (k) various low cost systems of public or private feedback can increase staff participation in their duties, (l) implementation of strict administrative policies can reduce tardiness and absenteeism, and (m) staff may be very reactive to announced observation by inspection agencies.