This study focuses on variation within a group of learners of Spanish (N = 253) who produced requests and complaints via a written discourse completion task. It examines the effects of learner and situational variables on production--the effect of proficiency and addressee-gender on speech-act choice and the effect of perception of imposition on request strategy. Unique to the design are 1) learners were not told what speech act to produce (but rather they chose how to respond); and 2) the design and sample size allowed for statistical techniques appropriate for nested data (regression in single level analyses and hierarchical linear models for nested data in multilevel models), which let one determine simultaneously the effects of learner and situation level variables on an outcome without violating the assumptions of the analytic technique. The following findings are described: more proficient learners produced requests more often, and less proficient learners produced complaints more often, net of other effects; female addressees were less likely than males to receive requests; and learners' perception of the imposition of the scenario did not significantly influence their production. The results highlight the need to account for learner and situational variables simultaneously when predicting speech-act production, using techniques appropriate for nested data.