Learners are often required to remember information about their (potential) learning partners, like their knowledge levels regarding several topics or their past participation level. Such so-called partner modelling processes help to make informed judgements about whom to select as a learning partner or to judge whether learning partners can provide help or need explanations. In two (pseudo-collaborative) experimental studies, we examined how partner models are formed in different computer-supported collaborative learning contexts. We applied multinomial processing tree models to measure memory processes unconfounded by guessing biases. In experiment 1 (N = 85), participants received information on the competence or participation level (high vs. medium vs. low) of 36 potential learning partners. High and low levels of participation were equally well remembered and better than medium levels. Regarding competence, high competence was better remembered than both, medium and low competence. In experiment 2 (N = 168), participants received information on the knowledge levels (high vs. medium vs. low) of a single learning partner (expert vs. intermediate vs. novice) regarding 60 topics. For the expert and intermediate partners, high and low levels were better remembered than medium levels. For the novice partner, low levels were remembered better than high and medium levels. Generally, the knowledge levels of the expert and novice were more accurately modelled than those of the intermediate partner. Findings support the general idea of context-dependent source memory regarding information about other persons and complement learning-related partner modelling research.