Instead of showing paired t test results, Prism gives error message 'A computation has divided a number by zero...".

A paired t test only makes sense when the two columns of data are strongly correlated. So as part of the paired t test calculations, Prism computes the correlation coefficient and associated P value.

If your data correlate perfectly, then r equals 1.00 and that P value cannot be computed. In that case, Prism reports "perfect line" as well as t test results. 

In rare cases, the correlation will be negative. As one variable gets bigger, the other gets smaller. In this case, it makes little sense to do a paired t test. If the correlation is perfectly negative, older versions of Prism gave an error message about dividing by zero, and gave no  t test results at all. This was fixed in 5.02 and 5.0b, which report  'perfect line' and give the full paired t test results.

This situation is rare, as it happens only when the data correlate perfectly,  so a graph of 'before' vs. 'after' forms a perfect straight line, with a negative slope. 

Here is a work around: Use the Remove baseline analysis to compute the difference between the two columns. Then use the Column statistics analysis to ask if the mean of those differences is distinct from 0.0. Those two steps are exactly what a paired t test does, so you'll get exactly the same results. But this way, you bypass the step where Prism tries to compute the correlation coefficient.  This file demonstrates. 

Another work around: If you were just playing around with made-up values (the usual situation when this bug appears), make up different values!