In settings where medications and viral load (VL) monitoring are limited by cost, clinicians need reliable ways to assess patient adherence to therapy. We assessed sensitivity and specificity of two self-reported adherence tools (a visual analogue scale [VAS] and the CASE [Center for Adherence Support Evaluation] adherence index), against a standard of detectable VL, with 288 patients from three sites in Thailand. We also assessed predictors of non-adherence. The sensitivity and specificity of the VAS <95% and CASE adherence index ≤11 against a VL >50 copies/mL were 26% and 90%, 19% and 95%, respectively. Against a VL ≥1000 copies/mL sensitivities increased to 55% and 36%, respectively, and specificities were unchanged. Attending a clinic not staffed by HIV specialists (odds ratio [OR] 3.14; 95% confidence interval [CI] 1.19-8.34) and being educated to primary school level or less (OR 2.24; 95% CI 1.01-4.94) were associated with self-reported adherence <95% on the VAS in multivariate analysis. Adherence assessed by the VAS was a more accurate predictor of detectable VL. Policy-makers in resource-limited settings should ensure that treatment centres are staffed with well-trained personnel aware of the importance of good patient adherence.