Shale pore space has a wide distribution of sizes (nm-μm) and complex configurations. Better knowledge of gas adsorption characteristics in real pore space is crucial for estimating shale gas-in-place. We develop a novel methodology to accurately and effectively calculate gas adsorption isotherms in multi-scale pore networks that simulate real pore structures inside shale. The influence of water saturations (in kerogen and clay) and pore distributions on gas adsorption is examined with our new model. 3D pore networks which connect both mesopores (2-50nm) and macropores (>50nm) are developed based on 2D SEM images and mercury intrusion analysis. Interparticle pores and pores inside kerogen have different morphologies from the pores in clay agglomerates in our pore networks. The gas adsorption on each dry pore/throat's surface is realized by capillary condensation with the Kelvin equation, which relates capillary condensation to pore/throat structure, different solid (clay and kerogen) surface characteristics and fluid properties. Moreover, we use the gas-liquid Gibbs adsorption model for gas adsorption on wet solid surfaces with water present, which is not considered in the literature. 3D pore networks and nitrogen adsorption isotherms are generated for the Silurian Longmaxi Formation shale samples. The simulated nitrogen adsorption isotherms are comparable to adsorption test results. The comparison confirms that both accurate adsorption modeling on pore surfaces and reliable pore space reconstruction are important for designing and analyzing adsorption measurements. Sets of methane adsorption isotherms are further calculated on different pore networks. Each pore network is assigned a unique combination of clay content, total organic carbon content and pore size distribution (PSD). When the pore volume is constant, shale has higher adsorption amount of methane with decreasing pore sizes. When the water saturation increases, water will first occupy the void space in clay from small pores to large pores and then extend to pores inside kerogen. It is concluded that the adsorption amount of methane could be significantly reduced by 50% when water saturation in pore space increases from zero to 30%. Different from previous adsorption modeling studies on single dry pore/throat or a bundle of dry tubes, this study considers the adsorption modeling on a pore network that connects pores and throats with different sizes, wet or dry surfaces and various morphologies. This methodology and simulation results are reliable and effective for fundamental study and field performance estimation of gas adsorption in shale reservoirs.