Many studies of data collection processes for business surveys focus on issues related to how to increase response rates and how to reduce response burden. Additionally, some have focussed on measurement error. Issues related to non response bias, on the other hand, do not seem to be explicitly part of the research agenda. The main reason why researchers should be concerned about non response is the potential for it to introduce bias. Our paper compares two alternative survey designs in terms of resultant response rates and non response bias. The first design is a simple postal survey with follow-up mailings; the second design is a two-phase multi-mode design, where the postal survey is followed at the second phase by a telephone survey of non-respondents. We found evidence that the sample obtained using only postal methods is biased in important respects. Bias is not apparent in the demographic characteristics of the employees. Bus bias is observed in some of the employees employment characteristics and some of the characteristics of the firms in which they work. The multi mode design seems, overall, to have reduced or removed the bias of the postal sample. Only in marginal respects was some further bias introduced. We also compare costs of the two designs, to enable a comparison of cost-effective at bias reduction.