This study investigates the impact of aspects of task design and implementation on oral second language performance in terms of task type, performance criteria and task-type repetition. One hundred and forty-four Chinese learners of English participated in the study. They were randomly paired and assigned to either a map task or a picture-story task. Two exemplars of each task type were used on separate occasions (one week apart). Each task group was further divided into two experimental groups - high criterion (HC) and low criterion (LC) - and a control group. Both the HC and the LC groups were informed at the outset that they would be asked to perform an additional activity upon completion of the baseline task. The He speakers undertook a more demanding post-task activity, and the LC group a less demanding one. No extra post-task requirement was imposed on the control group. A wide range of measures were employed to evaluate the participants' language production in terms of complexity. accuracy and fluency. A strong effect for task type was observed 011 all aspects of language performance, with the picture-story task giving rise to greater complexity both syntactically and lexically, and the map task generating more accurate and fluent language use. Overall. performance criteria were positively related to phrasal complexity and fluency on both types of tasks, but negatively associated with subordination complexity on the picture-story task. That is, in the case of the picture-story task, the HC group produced the highest phrasal complexity, but lowest subordination complexity, while the control group produced the opposite results. Such contrasting findings on complexity have not been reported in previous studies. In addition, repeating a similar task of the same type resulted in greater accuracy and repair fluency, whereas the temporal aspects of fluency deteriorated. The results are discussed mainly in terms of the limited attentional resources model on second language performance. The study contributes in particular to the current understanding of how pre-set task performance criteria can potentially affect language production. It also provides useful pedagogical implications.