Over the last decades, the inhibition of spontaneous burying of glass marbles by mice has been used as an index of anxiolytic drug action in the so-called marble burying test. Indeed, acute administration of rapid-onset (e.g. diazepam) and slow-onset (e.g. fluoxetine) anxiolytics inhibit marble burying. However, non-anxiolytic compounds such as classical antipsychotics also reduce marble burying thus suggesting that the predictive validity of this procedure for anxiety may be limited. In the present study, after having selected a strain of mice (C57BL/6J) that showed spontaneous avoidance of glass marbles, we tried to improve the predictive validity of the marble burying test for anxiety by measuring locomotor activity during the marble burying test and--if needed--in control experiments by using a videotracking system. Twenty-four reference compounds were tested including anxiolytics, anxiogenics, antidepressants, antipsychotics and other classes. By comparing marble burying scores with locomotor measures, we found that, based on our criteria, most of the anxiolytics and antidepressants selectively inhibited marble burying in contrast to most of the other compounds (e.g. haloperidol, morphine). Two putative anxiolytics, i.e. the nociceptin orphanin FQ peptide receptor agonist Ro 64-6198 and the metabotropic glutamate 5 receptor antagonist 2-methyl-6-(phenylethynyl)pyridine, also showed a selective profile. We propose this modified procedure, requiring only a limited number of animals, as a valuable screening test for the detection of compounds having anxiolytic effects.