Background In most countries throughout the world the construction industry continues to account for a disturbingly high proportion of fatal and nonfatal injuries. Research has shown that large enterprises seem to be most actively working for a safe working environment when compared to small and medium-sized enterprises. Also, statistics from Canada, Italy and South Korea suggest that the risk of injury among construction workers decreases with enterprise size, that is the smaller the enterprise the greater the risk of injury. This trend, however, is neither confirmed by the official statistics from Eurostat valid for EU-15 + Norway nor by a separate Danish study - although these findings might have missed a trend due to severe underreporting. In addition, none of the above mentioned studies controlled for the occupational distribution within the enterprises. A part of the declining injury rates observed in Canada, Italy and South Korea therefore might be explained by an increasing proportion of white-collar employees in large enterprises. Objective To investigate the relation between enterprise size and injury rates in the Danish construction industry. Methods/Design All male construction workers in Denmark aged 20-59 years will be followed yearly through national registers from 1999 to 2006 for first hospital treated injury (ICD-10: S00-T98) and linked to data about employment status, occupation and enterprise size. Enterprise size-classes are based on the Danish business pattern where micro (less than 5 employees), small (5-9 employees) and medium-sized (10-19 employees) enterprises will be compared to large enterprises (at least 20 employees). The analyses will be controlled for age (five-year age groups), calendar year (as categorical variable) and occupation. A multi-level Poisson regression will be used where the enterprises will be treated as the subjects while observations within the enterprises will be treated as correlated repeated measurements. Discussion This follow-up study uses register data that include all people in the target population. Sampling bias and response bias are thereby eliminated. A disadvantage of the study is that only injuries requiring hospital treatment are covered.