You don't mention which of the MEME Suite tools you were using. I'm guessing it was either STREME or SEA. Both of these tools compare the number of motif matches in the target set with the number of motif matches in a control set in order to evaluate the statistical significance of the motif or of its enrichment in the target set. If you aren't providing a control set STREME and SEA create one by shuffling the target set.
For the statistics to work though, the target and control sets have to have similar length distributions. That is, if the target set has an average length of 500, and the control set has an average length of 5000, the estimation of statistical significance won't be reliable. If the imbalance is big enough SEA and STREME will quit. There can also be problems when most of the sequences of the target set are around length 500, but a few are of length 10,000 or length 10. In general you want the sequences to be of roughly the same length, at least say within the same order of magnitude.
You should review your sequences and make sure they look roughly consistent in length.