Using data from the Household, Income and Labour Dynamics in Australia (HILDA) Survey, we find evidence that unobserved heterogeneity substantially biases cross-sectional estimates of union wage effects upward for both male and female workers. The estimate of the union wage premium for male workers falls from 8.7 per cent to 5.2 per cent after controlling for unobserved heterogeneity. For females the estimated 4.0 per cent cross-sectional union wage premium falls to 2.0 per cent once unobserved heterogeneity is controlled for. Among males, the results suggest negative sorting into union status at high levels of observed skill, while the sorting is positive at the lower end of the skill hierarchy. There is not sufficient evidence to support a similar pattern of sorting among female workers.