A problem that frequently occurs in metrology is the bias checking of data obtained by a laboratory against the specified value and uncertainty estimate given in the certificate of analysis. The measurand-a property of a certified reference material (CRM)-is supposed to be random with a normal distribution whose parameters are given by the certificate specifications. The laboratory's data from subsequent measurements of the CRM (a CRM experiment) are summarized by the sample mean value and its uncertainty which is commonly based on a repeatability standard deviation. New confidence intervals for the lab's bias are derived. Although they may lack intuitive appeal, those obtained by using higher order asymptotic methods, compared and contrasted in this paper, are recommended.