Classification of pig carcasses in the European Community is based on the lean meat percentage of the carcass. The lean meat percentage is predicted from instrumental carcass measurements, such as fat and muscle depth measurements, obtained in the slaughter-line. The prediction formula employed is derived from the data of a dissection experiment and has to meet requirements for authorization as put down in EC regulations. Requirements involve the sampling procedure and sample size for the dissected carcasses and the accuracy of prediction. Formulae are often derived by linear regression. In this paper we look at a particular type of sampling scheme. This involves selection of carcasses on the basis of carcass measurements not all of which are intended to be used as prediction variables. This sampling scheme frequently appears in requests for authorization of carcass measurement instruments and accompanying prediction formulae, despite the fact that it lacks formal statistical justification when used in conjunction with linear regression. The objective of this work was to assess the performance of the prediction formula that follows from this potentially faulty combination of sampling scheme and linear regression in relation to the requirements in the EC regulations. We show that this sampling scheme may produce poor predictions for lean meat percentage compared with proper sampling procedures with selection on prediction variables only or random sampling. We do so by computer simulation. Initially, simulated data were based on recent and historic data from The Netherlands. Prediction variables are fat and muscle depth measurements. The additional variable involved in sampling, but not included in the regression, was carcass weight. We also show that due to this faulty sampling scheme there is a serious risk that a new measurement instrument may not be authorized because performance criteria in the EC-regulations are not met.