Chatbots have shown potential in language learning. The purpose of the research was to explore the effects of SimBot, a virtual chatbot with ASR rating feature, on EFL elementary school students’ English pronunciation. A total of 23 students practiced English speaking with SimBot individually in meaningful conversation scenarios once a month for four months. Students’ dialogue recordings were rated by researchers in four categories including vowel, consonant, word stress, and overall intelligibility. Students’ pronunciation errors were further classified into four categories namely consonant dropping, consonant substitution, over correction, and vowel mispronunciation. Each student’s pronunciation errors were tabulated to observe students’ pronunciation learning trajectory. Wilcoxon signed-rank test showed that participants improved in the consonant category. Moreover, of the four errors category, only consonant substitution did not improve. The researchers speculated that it might be difficult for ASR to detect the differences between consonant sounds such as the liquid consonant /l/ and /r/ (the major errors in consonant substitution), thus resulting in students not noticing errors when they practiced speaking with SimBot. This research revealed English pronunciation errors that commonly made by Taiwanese EFL elementary school students and pointed out the limitation of using ASR in English pronunciation learning.