Online surveys are becoming the dominant form for survey data collection. This presents a problem for the measurement of political knowledge, because, according to recent scholarship, unsupervised measurement of political knowledge in web-based surveys suffers from respondent dishonesty. This study examines the validity of five possible survey proxies for political knowledge: self-assessed sophistication, political interest, internal political efficacy, accuracy of party placements on a left-right dimension and political participation. The analysis draws on a 2020 survey data (n = 1,097) and partial replications with identical measures from a 2008 survey data (n = 1,021) from Finland. Through several tests, the five proxies are assessed in terms of convergent validity, criterion validity and predictive validity. Across all tests, political interest performs best on all dimensions of validity and demonstrates largely identical relationships with political knowledge. Although the survey measurement of political interest and political knowledge may partly tap into slightly different constructs, the analysis supports the conclusion that political interest is the most suitable survey proxy for political knowledge from among the five proxy candidates included in the analysis.