Telemedicine’s dramatic increase during the COVID-19 pandemic elevates the importance of addressing patient-care gaps in telemedicine, especially for patients with limited English proficiency. To examine the associations of patient language and patient-provider language concordance with telemedicine visit type (video versus telephone visit). Cross-sectional automated data study of patient-scheduled primary care telemedicine appointments from March 16, 2020, to October 31, 2020. Northern California integrated healthcare delivery system. All 22,427 completed primary care telemedicine visits scheduled by 13,764 patients with limited English proficiency via the patient portal. Cross-sectional association of electronic health record-documented patient language (Spanish as referent) and patient-provider language concordance with patients’ choice of a video (versus telephone) visit, accounting for patient sociodemographics, technology access, and technology familiarity factors. Of all patient-scheduled visits, 34.5% (n = 7747) were video visits. The top three patient languages were Spanish (42.4%), Cantonese (16.9%), and Mandarin (10.3%). Adjusting for sociodemographic and technology access and familiarity factors and compared to patients speaking Spanish, video visit use was higher among patients speaking Cantonese (OR = 1.34, 95% CI: 1.18-1.52), Mandarin (OR = 1.33, 95% CI: 1.16-1.52), or Vietnamese (OR = 1.27, 95% CI: 1.09-1.47), but lower among patients speaking Punjabi (OR = 0.75, 95% CI: 0.75, 0.62-0.91). Language concordance was associated with lower video visit use (OR = 0.86, 95% CI: 0.80-0.93) and moderated associations of speaking Spanish, Cantonese, and Korean with video visit use. In addition, for all language groups, those with prior video visit use were more likely to re-use video visits compared to those with no prior use (p < .05 for all languages except Hindi with p = 0.06). Among linguistically diverse patients with limited English proficiency, video telemedicine use differed by specific language. Disaggregating patient subpopulation data is necessary for identifying those at greatest risk of being negatively impacted by the digital divide.