The recent technology improvements in CryoEM which have been responsible for the rapid growth in high-resolution structures has also led to comparable improvements in electron cryotomography as well as the study of conformational and compositional heterogeneity in single particle analysis. The improved contrast in each single particle image permits more accurate classification of individual particle images, and hence allows a finer gradation of structural variations to be studied both qualitatively in 2-D as well as quantitatively in 3-D. We present several approaches for these problems applied to proteins with a mix of ordered and disordered domains. In CryoET, improved contrast has permitted separation of smaller objects accurately from the crowded cytosol of 3-D reconstructions of cellular material. Previously it was challenging to isolate objects much smaller than the ribosome, but now sub-megadalton particles are becoming separable. The most challenging problem in processing such cellular tomograms has historically been the segmentation process wherein recognizable structures are identified among the ~10 billion voxels in a high resolution tomogram. This process is highly labor intensive with as much as a man-week required for complete analysis of a single cellular tomogram, even with the best assistive tools currently available. We present a new neural-network based semi-automatic classifier, which is trained to mimic a human annotator with a very high degree of reliability, dramatically reducing the human effort required for the cellular segmentation process. We also discuss new tools for subtomogram averaging to achieve higher resolution structures of complexes in-situ and compensating for missing wedge artifacts.