It is commonly argued that socio-economic inequalities can explain many of the differences in achievement and participation in science education that have been reported among countries and among schools within a country. We addressed this issue by examining (a) the relationship between variables associated with socio-economic background and application frequencies to the Swedish Natural Science Programme (NSP) in upper secondary school and (b) whether there are lower secondary schools in Sweden that seem to compensate for these variables. Data from Statistics Sweden (SCB) covering the whole population of 106,483 ninth-grade students were used to calculate the probability for each student to apply to the NSP. Our results indicate that the variables, such as parental educational level and grades, have explanatory power, but with varying effect for different subpopulations of students. For example, grades in mathematics have a greater impact than grades in science for females' choice of the NSP. The opposite holds for male students. Out of 1,342 schools, 158 deviated significantly from predicted, that is, the students in these schools applied to the NSP in greater or lesser extent than expected. The number of deviating schools is greater than predicted by pure random variation. This suggests that variables of socio-economic background are only a partial explanation of the application frequencies, and that the deviation needs to be investigated further. Our findings suggest that in order to understand why schools deviate positively and so compensate for the socio-economic background of their students, we need to study their practices more closely.