It means that your test set is not independent of your model. If you solved this structure by MR using a closely related model, you should use the same test set for the new structure, extending it to higher resolution if appropriate. The test set can be transferred using data manipulation utilities in phenix or ccp4.
If it's not possible to maintain the same test set, for example your new crystal is in a different space group, then simulated annealing of your model can help remove model bias. You'll probably have to do quite a bit of rebuilding to remove errors introduced by the simulated annealing.
Best wishes
Kevin