I am running local refinement coupled with masked 3D classification on a target that refines to around 3 Å resolution in the core. I have an area of interest that I would like to resolve better, so I ran local refinement on that region, then 3D classified, then repeat this whole cycle a second time to arrive at a selection of particles. What I’m puzzled about is the very low B-factor I’m getting out from the local refinements. After the first cycle the best class has a B-factor of around 30, and the second cycle outputs B-factors <20. I get an estimated resolution of 3.5 Å from 16k particles which seems reasonable to me with such a low B-factor. The consensus refinement has a B-factor of around 90 with close to 1M particles. I am not doing any particle subtraction. Is this expected behavior?