Kelly,
Latent class assigns each respondent probabilistically to each segment. So a segment that has 0.4 of respondent 1 and 0.9 of respondent 2 and 0.01 of respondent 3 will report in the software as having 1.31 respondents. The software also reports for each respondent their modal segment, and if that's what you imported to SPSS, then the two (counts of whole respondents put in modal segments and sums of respondent-level segment probabilities) will usually disagree a little. It's not that they're counting the same thing two different ways, but that they're different sums.