Abstract This dissertation discusses the effect of intra-task variations in association with performance conditions on task difficulty in a semi-direct speaking test, namely, the GEPT-I. The variables identified for investigation are linguistic demand of task input in relation to code complexity; amount of time allowed for performance in relation to communicative demand; type of pre-task planning in relation to communicative demand; and test-takers’ familiarity with the non-verbal propositional content of the task input in relation to cognitive complexity. The dissertation takes a multi-dimensional approach in measuring the effects of the variables in a comparison of data collected from the performances in a controlled task and in an experimental task. The dimensions of the comparison are based on three different sources of data, including task scores, test-takers’ responses to the post-task questionnaires eliciting views of difficulty, and interlanguage measures in the areas ...