As response rates to surveys decline all over the world, researchers are increasingly turning to sampling frames that are easier and cheaper to reach, and that have more predictable response rates. These include nonprobability web panels (NWPs) and probability web panels (PWPs). Although generally more expensive to construct, the latter have been shown in many instances to suffer from fewer biases and deviation from benchmarks. The literature comparing NWPs with PWPs is fedgling. We add to this research area by comparing measures of the social determinants of health that were estimated from a number of NWPs and PWP equivalents with a high-quality benchmark. The analysis finds that, when looking at the distributions of self- assessed health and life satisfaction, probability panels differ less from the gold standard than do nonprobability panels. This supports previous work, although we also show that this conclusion holds when a greater range of control variables is included in the model. However, some of the predictors of health are captured better using the nonprobability panels. In particular, the relationship between area-level disadvantage and health is better captured through a pooled nonprobability sample.