XFIT gives a weighting of some type on each point (1/(X^2)?), thus if a count is zero - XFIT detects that a potential 1/0 problem could occur and refuses to refine. In the case of a random point, change the 0 count to 1. If in the case of energy dispersive data, which can have large arrays of zero counts, one possibility is to add 1 to every channel so this problem does not occur. If all your counts are above zero, one thing to consider is that you have converted your raw data incorrectly and it lacks a point or two. Don't forget that if you have data going from 20 to 80 at 0.02 steps, you do not have 3000 points ((80-20)/0.02) but 3001 points due to the starting point ((80-20)/0.02)+1.