Radiographic inverse problems are often treated deterministically, while probabilistic approaches for uncertainty-aware inference remain comparatively underdeveloped. In this work, we present Bayesian methods for multimodal radiographic inverse problems, focused on quantitative material-thickness estimation across a range of settings with increasing complexity, from synthetic studies to simulation-based data to experimental laboratory X-ray and neutron radiography data. Our framework uses Markov chain Monte Carlo based inference with geometrically reparameterized inputs to respect physical constraints and improve posterior exploration, and it incorporates an embedded model error formulation to represent uncertainty arising from model-form limitations in addition to measurement noise. The resulting approach enables quantitative material reconstructions with principled uncertainty estimates, illustrating the value of Bayesian probabilistic methods for robust radiographic inference.