You can use any of the squares - the 9 large squares are all the same size. The large and small subdivisions are to help with different sizes of cells and different concentrations of cells. For example, to count very small cells like RBC it would be easier to use higher magnification so you would use a more concentrated suspension and focus on the central square with the most subdivisions. Our rule is to count enough large squares to get a total count of at least 150 cells, and then to do that again with 2 further independently picked samples from the original cell suspension (= 3 in all) and make an average, if you want an accurate count.
Don't forget to re-mix your suspension before taking a sample, as cells settle down to the bottom slowly,
We also keep the cells on ice while counting, as some cell types can start attaching to the container at room temperature.