Vitkus and Horowitz (1987) found that lonely people demonstrated adequate social behavior when they were assigned to controlling interpersonal roles. Despite this successful performance, they evaluated themselves and their behavior negatively. Study 1 replicated these findings and extended them to naturalistic interactions. In a hypothetical problem-solving dyad involving paired college students (N=76), students were diagnosed as either lonely or nonlonely and then assigned respective roles of Person with the Problem or Springboard (listener/advisor). Analysis of videotaped sessions showed that all subjects were capable of adequate social performance, but adopting a passive social role interfered with expression of this behavior. Study 2 followed the procedures of Study 1 with the difference that all subjects were assigned to the Springboard role (who listens to a partner describe the problem) and positive or neutral feedback was provided. Results showed that unambiguous positive feedback reduced the negativity of lonely students' self-evaluations. An interpersonal model of loneliness is proposed to explain these results. The model suggests that in typical interactions, lonely people adopt social roles that prevent them from expressing appropriate social behavior. In addition, without clear performance cues, lonely people fail to recognize occasions when they do perform adequately. Their resulting negative self-appraisals reaffirm their initial feelings of inadequacy, thereby sustaining their loneliness. Therapeutic implications of this model and limitations of the present research are discussed. (TE)