While numerous empirical studies include proxies for growth opportunities in their analyses, there is limited evidence as to the validity of the various growth proxies used. Based on a sample of 1,942 firm years for listed UK companies over the period 1990 to 2004, we assess the performance of eight growth opportunities measures. Our results show that while all the growth measures show some ability to predict growth in company sales, total assets or equity, there are substantial differences between the various models. In particular, Tobin's Q performs poorly while dividend based measures generally perform best. However, none of the measures has any success in predicting earnings per share growth, even when controlling for mean reversion and other time-series patterns in earnings. We term this the 'growth companies puzzle'. Growth companies do grow, but they do not grow in the key dimension (earnings) theory predicts. Whether the failure of 'growth companies' to deliver superior earnings growth is attributable to increased competition, poor investments, or behavioural biases, it is still a puzzle why growth companies on average fail to deliver superior earnings growth.