This mixed-methods study investigated English-as-a-foreign-language (EFL) learners’ perceptions of task difficulty and their use of metacognitive strategies in completing integrated speaking tasks as empirical evidence for the effects of metacognitive instruction. A total of 130 university students were invited to complete four integrated speaking tasks and answer a metacognitive strategy inventory and a self-rating scale. A sub-sample of eight students participated in the subsequent interviews. One-way repeated measures MANOVA and structure coding with content analysis led to two main findings: (a) EFL learners’ use of metacognitive strategies, in particular, problem-solving, was considerably affected by their perceptions of task difficulty in completing the integrated speaking tasks; (b) EFL learners were not active users of metacognitive strategies in performing these tasks. These findings not only support the necessity of taking into account learners’ perceptions of task difficulty in designing lesson plans for metacognitive instruction, but also support a metacognitive instruction model. In addition, the findings provide empirical support for the utility of Kormos’ Bilingual Speech Production Model. As the integrated speaking tasks came from a high-stakes test, these findings also offer validity evidence for test development in language assessment to ascertain sustainable EFL learning for nurturing learner autonomy as an ultimate goal.