Characterizing the pore structure of shale is essential to understanding fluid transport through the matrix and optimizing any stimulation plan. Organic shales are heterogeneous at multiple scales, and the characteristic length scales or correlation lengths are often longer than the scale of samples used for laboratory analysis. Using laboratory data to make predictions at the wellbore scale therefore requires careful upscaling. Using samples of Barnett and Eagle Ford shales, and a siliceous, oil-bearing shale from the northern Rocky Mountains, we performed high- pressure mercury intrusion (HPMI) and low-pressure nitrogen sorption. We determined the properties of the pore network (size distribution, connectivity, spatial correlation) by constructing representative pore networks that allowed reconstruction of the HPMI and nitrogen sorption data. We then upscaled the results determining the correlation length with a percolation-based scaling function. Based on the HPMI and nitrogen sorption measurements, pore networks tend to be very poorly connected at the micron scale, with average coordination numbers between 2 and 3. Clusters of connected pores are typically a few hundred microns in size. Our work has significant implications for using laboratory measurements to predict reservoir properties. While samples are relatively homogeneous at the scale of SEM images or HPMI/nitrogen sorption measurements, organic-rich samples in particular have longer-range correlations that are not captured at this scale and yet exert significant control on transport properties. This will affect production from a fracture-stimulated well since induced microcracks and their interactions with the in situ pore structure are extremely important for moving hydrocarbons toward the main induced fracture system, as demonstrated by previous researchers. Multi-scale characterization is therefore necessary to gain a full understanding of the shale matrix.