the long-term selection experiments allow to study the limits for the response to artificial selection and the consequences in other unselected characters. the selected lines used in this type of experiments, generally, had a ne>40 and they are originated from outbreed populations with a wide genetic basis. nevertheless, in 1986, at the faculty of veterinary sciences, national university of rosario, argentina, two lines of two-way selection for body weight were founded with animals of an unselected cf1 mice population with ne<40. the theoretical estimate of the initial genetic variance of the founders was closed to zero because of the cf1 strain, like anyone strain, was originated from 20 generations of full sibs matings. the results of 60 generations of selection are presented. the consequences of the selection in other characters related with fitness and with production were also studied. the response to selection demostrated the conservation of genetic variance in the founders and the presence of non-additive genetic components. the fitness deterioration of the selected lines led to a decrease in the production efficiency when they are compared with the non-selected line. in the long-term, selection for weight allowed to obtain heavier individuals but fitness was reduced. this implies a high selection cost that translates in a low production when the process is evaluated at a population level. a modification to the lerner's model of genetic homeostasis with greater fitness for the intermediate phenotypes was proposed to explain these results.