Aim/Purpose: This study examines the effects of the mobile-assisted task-based language teaching (M-TBLT) approach on EFL learners' oral production. It evaluates three key second language acquisition measures: complexity (syntactic and lexical), accuracy (error-free clauses and correct verb forms), and fluency (unpruned and pruned speech rates). Additionally, it explores learners' perceptions of the approach to gain deeper insights into its effectiveness. Background: Task-Based Language Teaching (TBLT) provides a communicative framework but often faces challenges such as insufficient oral practice, delayed feedback, and limited authenticity. Mobile-assisted language learning (MALL), with its flexibility and real-time interaction, offers potential solutions. However, empirical research integrating MALL with TBLT in oral English instruction remains scarce. This study introduces a structured six-step speaking instruction model to assess the impact of M-TBLT on learners' oral performance. Methodology: A quasi-experimental pretest-posttest control group design was employed. Participants were university students from two intact speaking classes in Hebei, China. The intervention incorporated two mobile applications--WeChat and Liulishuo--within a structured six-step speaking instruction model. Data collection combined quantitative measures (pretest-posttest assessments) with qualitative insights from interviews and a teacher journal. Contribution: This study fills a research gap by providing empirical evidence on the integration of mobile applications with TBLT in oral English instruction. Unlike existing studies that focus on spontaneous and unstructured speaking activities, this research integrates a well-sequenced instructional model. Findings contribute to both theoretical discussions and practical applications in mobile-assisted language learning. Findings: Results indicate that M-TBLT significantly improved students' speaking complexity (dependent clauses per T-unit), accuracy (error-free clauses and correct verb forms), and fluency (unpruned and pruned speech rates). However, no significant improvement was observed in lexical complexity (mean segmental type-token ratio). Qualitative findings reveal that participants had positive perceptions of M-TBLT, appreciating mobile affordances (ease of use, synchronous/asynchronous communication, repetition, instant feedback, and authenticity) and educational benefits (increased speaking practice and learning autonomy). Students particularly valued WeChat's voice recording and group functions, as well as Liulishuo's automatic speech recognition and shadowing features. Recommendations for Practitioners: The structured speaking-instruction model offers educators a practical framework for implementing M-TBLT, enhancing both instructional design and students' speaking performance. Recommendation for Researchers: This study provides empirical support for M-TBLT in EFL settings. Researchers may explore its application in diverse linguistic and cultural contexts and examine additional instructional variables. Impact on Society: These findings benefit educators, researchers, and mobile technology developers, encouraging collaboration in advancing mobile-assisted task-based learning. Future Research: Further research could investigate M-TBLT in varied learning environments, incorporate additional performance measures, and assess new mobile-assisted speaking tools to enhance task-based learning.