Background: Poorer virologic response to NVP vs EFV-based ART has recently been reported. We compared VL response in ARROW trial children aged ≥3 years initiating ART with NVP vs EFV. Methodology: 836 ART-naive children age ≥3 years from Uganda/Zimbabwe were randomized to 3-drug NNRTI+3TC+ABC (Arm A) vs 4-drug NNRTI+3TC+ABC+ZDV induction, decreasing at week 36 to 3-drug NNRTI+3TC+ABC (Arm B) or 3TC+ABC+ZDV (Arm C). NNRTI was chosen by clinicians. VL was assayed using Abbott Taqman or Roche Amplicor, with lower detection limit 80 c/ml. Predictors of suppression <80 c/ml at 36wk (all children) and 144wk (Arm A/B only) after ART initiation were identified using backwards elimination (p=0.1) from pre-ART CD4%, weight/height-for-age, CD4%, VL, sex, centre, ART randomization, CD4 monitoring, and reported missed ART doses in the last 4wk, forcing EFV vs NVP and age at ART initiation into the model, and categorizing age to reflect non-linear effects. Results: 445 (53%) children received EFV and 391 (47%) NVP. Children on EFV were more likely to be male (58%, p=0.007), older (median 8.6 year EFV vs 7.5 NVP, p<0.001), less underweight (weight-for-age -1.7 vs -2.4, p<0.001) and had higher CD4% (12% vs 10%, p=0.05) and lower pre-ART VL (160011 vs 203882c/ml, p=0.07), but similar pre-ART CD4 (p=0.9). 631 (75%) had ≥1 post-baseline VL. Overall there was a trend to better suppression <80 c/ml with EFV (Figure; global unadjusted p=0.08, particularly ≤36wk). After adjustment, at wk 36 (n=274 to date), EFV was superior to NVP in those with pre-ART VL>75000 (p=0.001; interaction p=0.003); suppression <80 c/ml was poorer in the youngest and oldest children (aOR(3-4 vs 5-9yr olds)=0.57 p=0.14; aOR(10+ vs 5-9yr olds)=0.33 p=0.007). Differences at wk144 were smaller (75% EFV vs 70% NVP <80 c/ml (n=196)) but still significant (p=0.03), with a strong relationship between older age and poorer VL suppression in those on EFV (aOR per year older=0.78 [0.68-0.89] p<0.001) but no relationship in NVP (aOR =0.97 [0.80-1.17] p=0.74) (age interaction p=0.06). Conclusions: In a non-randomized NNRTI comparison, although short-term VL suppression favoured EFV, long-term suppression depended on age: children <10 years at ART initiation had better long-term VL suppression with EFV than NVP; whereas those >10 years had better long-term suppression with NVP than EFV. The relative performance of EFV vs NVP therefore varies over time on ART, by baseline VL and by age at ART initiation, the latter possibly due to CNS side-effects of EFV.