There are two major variants of the chi-square test (for frequencies in tables). The chi-square test of independence is perhaps the most common, and Charlotte Eost-Telling
's explanation will help you to elicit that version. This is appropriate when your hypothesis concerns whether two categorical (or higher scale) variables are independent (vs. the alternative that they are related).
The chi-square goodness of fit is the other version, and is pertinent when the hypothesis has to do with questions of whether an observed distribution of frequencies in a categorical (or higher scale) variable match well to some expected / theoretical / hypothesized distribution. That version is found in the non-parametric section of spss, and can be activated as Mohammad Mohaiminul Islam 's suggests.