Excluding exposure data of very poor quality is a core principle for regulatory risk assessment.

In a recent issue of this journal, Money and Margary (2002) proposed some thoughtful guidelines for exposure assessment in regulatory risk assessment. Their structured approach acknowledges that currently available exposure data are of very variable quality (Northage and Marquart, 2001). We firmly underline their plea for a hierarchy in available exposure information sources, with a higher weight assigned to data with a lower level of uncertainty. The level of confidence an assessor has in the available exposure information should play an important role in the resultant risk assessment process. Moreover, the classification of data into different uncertainty categories should aid the assessor when conflicting exposure results are reported. Hence, a transparent system quantifying heterogeneity in data quality is crucial if one wants to arrive at consistent risk assessments. The approach of Money and Margary largely coincides with and was to some extent complementary to our decision tree for data quality evaluation published in the same issue (Tielemans et al., 2002). However, we question their statement that all exposure information sources should be considered as being potentially useful in the risk assessment process. On the contrary, it is our opinion that not all exposure information meets even the minimum requirements for incorporation in the exposure assessment process and the exclusion of such data should be the starting point for a transparent and robust exposure assessment. In our paper we defined minimum requirements for four different aspects, i.e. available occupational hygiene information, variability and precision issues, internal validity and external validity. We consider data sources to be unacceptable if very basic requirements are not fulfilled for these aspects. In these cases, the level of uncertainty or bias related to exposure data is in our view difficult to interpret in even a broad sense. In order to evaluate the quality of data in current European Union risk assessments of existing substances, we conducted a small-scale inventory of exposure data. One exposure assessor of our department evaluated data quality of 40 measurement series selected out of five Risk Assessment Reports (RARs). A second researcher also evaluated 20 of these sources in order to study agreement between assessors. The data classification was done according to both a strict and a lenient interpretation of our decision tree (Tielemans et al., 2002). The former implies a rigid adherence of the assessor to the rules of the decision tree. Any non-compliance to the decision rules results in exclusion of the data. The latter refers to an approach that leaves more room for subjective assessment tailored to the specific exposure assessment situation. Table 1 describes the results of our inventory. It is a striking finding that 80% of the information sources were excluded when a strict classification was applied. A lenient approach also yielded exclusion of several measurement series (12.5%), although most ratings shifted towards supplementary information (80%). It should be noted that in both approaches only a small percentage of sources resulted in sufficient information. The percentage agreement between both assessors was 85 and 70% for a strict and lenient approach, respectively. This difference is in accordance with expectation, since a less rigorous approach relies to a larger extent on subjective judgement. The outcome of our analysis clearly illustrates the contrast between a rigorous evaluation that dismisses most data as useless and a lenient approach that allows most data to be incorporated into the risk assessment process. It can be learned from this inventory that a large part of the investigated data is hampered by a poor documentation of occupational hygiene information, low precision or questionable validity. A scientifically rigorous assessment would classify these data as being of very little relevance. We agree with Money and Margary that, as a result of