This study compared the language produced by test takers, as well as the holistic rating scores between semi-direct (human-to-machine) and video-conferencing modes. We used the Standard Speaking Test (SST) for the video-conferencing mode and the Telephone Standard Speaking Test (TSST) for the semi-direct mode. Our results indicate that examinees received better holistic rating scores in the semi-direct mode than in the video-conferencing mode. Analytical measures revealed that the semi-direct mode exhibits superior syntactic complexity at the expense of accuracy and fluency.