This all depends on what kind of image you have loaded. The threshold levels will represent t-values if you have loaded a student-t image. Any student-t value below or above the min/max t-values given will then be invisible. You need to convert the image from student-t to p-values if you want the threshold levels to threshold on p-values. Some tools, like FSL's randomise, will give you p-valued images and then you will threshold on the p-values.