Background We sought to explain the variation in injury rates found for categories of companies and contracts involved in the construction of the Denver International Airport (DIA) by surveying contractors about company and contract-level safety practices. Methods We conducted 213 telephone interviews (83% response) with representatives of contracts with payrolls of more than $250,000. We investigated the bivariate relationship between safety actions reported in the survey and injury occurrence by calculating the aggregate injury rates (lost work-time (LWT) rates and non-LWT rates) for the group of respondent contracts reporting always taking the action and for the group not always taking the action. Using Poisson regression, we examined the association between contract injury rates and contract safety practices while controlling for variables previously shown to affect contract-level injury rates. Results In Poisson regression, two actions, 1) disciplinary action always resulting when safety rules were violated and 2) always considering experience modification ratings when selecting subcontractors, were associated with lower LWT injury rates. Three actions or contract characteristics resulted in lower non-LWT rates: management always establishing goals for safety for supervisors, conducting drug testing at times other than badging or after an accident, and completing the DIA contract on budget, rather than over budget. Reportedly consistent use of a number of accepted safety practices was associated with significantly higher injury rates in bivariate and multivariate analyses. Conclusions The pattern of counterintuitive results found in this study suggests that questions reflecting agreed-upon safety practices, when asked of the person responsible for all on-site construction activities, are likely to elicit normative responses. Objective validation of reported safety practices is critical to evaluating their effฤฑcacy in reducing injury rates, along with measures of both time at risk and outcome and control for prevailing risk of the work performed.