Realistic modeling of dense pre-graphitic carbons is a prerequisite for the fine understanding of their properties. High Resolution Transmission Electron Microscopy (HRTEM) allows to get valuable 2D information about the atomic structure of such carbons. However, HRTEM is 2D by nature while the observed structures are 3D. Obtaining 3D representations of such materials would be a great help in the process of full 3D atomic structure modeling. We propose an approach which aims at synthesizing 3D image data from 2D HRTEM images using a novel analysis/synthesis framework. The analysis step consists in computing reference statistics on the subbands of a HRTEM sample decomposed using a steerable pyramid. In the synthesis stage, through an iterative and multiscale scheme, these statistics are imposed in 3D on the subbands of a random 3D data block. The inference from 2D reference statistics to 3D target statistics is done using the property of orthotropy typical of pre-graphitic carbons. Synthesis results obtained from HRTEM samples with increasing structural order are provided which show the relevance of the synthesis scheme.