We use a randomized field experiment to estimate the effect of computer-assisted personal interviewing (CAPI) on survey errors, which are grouped into four categories: logical, skip, validation, and missing errors. During the 2017 Agricultural Household Survey for Anuradhapura district, Sri Lanka, 1910 households were randomly selected and interviewed using CAPI, while another 1910 households were randomly selected and interviewed using pencil-and-paper interviewing (PAPI). We find that CAPI interviews had on average 36 less errors per questionnaire, with missing errors being the largest contributor. We also find that household questionnaire errors are lower, and it is the operator questionnaire errors that are driving the results. Finally, CAPI really helps with survey-to-survey matching of parcels, but within-survey matching gives us mixed results.