The availability of pluripotent embryonic stem (ES) cells for gene targeting has resulted in laboratory mice becoming important animal models of human neurological disease. Inbred strains of mice differ in many behavioural phenotypes, such that the same gene mutation can appear to have different phenotypic effects when introduced onto different genetic backgrounds. Prior knowledge of the behavioural phenotypes of the inbred strains used for gene targeting would, therefore, allow the selection of the most appropriate genetic background for the hypothesis to be tested. With this in mind, we tested eight strains of mice (129S1/SvImJ, 129S2/SvPasIcoCrlBR, 129S6/SvEvTac, B6129SF1/J, C57BL/6J, C57BL/6N, LP/J and SM/J), including the sources of five ES cell lines commonly used for gene targeting, in the spatial (submerged platform) version of the Morris water maze, the most widely used paradigm to evaluate the cognitive abilities of genetically modified mice. The three 129 substrain sources of ES cell lines demonstrated spatial learning in the water maze that was superior to that of C57BL/6J, the inbred strain most commonly used for the maintenance and phenotypic testing of mutations. In addition, 129S6/SvEvTac was unique amongst the eight strains tested in having a particular capacity for reversal learning, when the submerged platform was relocated to the opposite quadrant. We conclude that some substrains of 129 could provide suitable genetic backgrounds for testing gene mutations that might be expected to impair cognitive function, thus negating the need to backcross to C57BL/6J, thereby avoiding the so-called "flanking gene problem".