Objectives: The aim of the study was to assess the prevalence, predictors and patterns of genotypic resistance mutations in children after failure of World Health Organization-recommended initial nonnucleoside reverse transcriptase inhibitor (NNRTI)-based treatment regimens. Methods: We carried out a multicentre retrospective study of genotyping tests performed for all HIV-infected children at eight paediatric centres in Thailand who experienced failure of NNRTI therapy at a time when virological monitoring was not routinely available. Results: One hundred and twenty children were included in the study. Their median age (interquartile range) was 9.1 (6.8-11.0) years, the median duration of their NNRTI regimens was 23.7 (15.7-32.6) months, their median CD4 percentage was 12% (4-20%), and their median plasma HIV RNA at the time of genotype testing was 4.8 (4.3-5.2) log10 HIV-1 RNA copies/mL. The nucleoside reverse transcriptase inhibitor (NRTI) resistance mutations found were as follows: 85% of the children had M184V/I, 23% had at least four thymidine analogue mutations, 12% had the Q151M complex, 5% had K65R, and 1% had the 69 insertion. Ninety-eight per cent of the children had at least one NNRTI resistance mutation, and 48% had etravirine mutation-weighted scores ≥4. CD4 percentage <15% prior to switching regimens [odds ratio (OR) 5.49; 95% confidence interval (CI) 2.02-14.93] and plasma HIV RNA>5 log10 copies/mL (OR 2.46; 95% CI 1.04-5.82) were independent predictors of at least four thymidine analogue mutations, the Q151M complex or the 69 insertion. Conclusions: In settings without routine viral load monitoring, second-line antiretroviral therapy regimens should be designed assuming that clinical or immunological failure is associated with high rates of multi-NRTI resistance and NNRTI resistance, including resistance to etravirine. © 2010 British HIV Association.