This work employs a multiscale modeling approach to probe the thermal conductivity in sintered randomly oriented carbon nanotube (CNT) networks, in which a fraction of close contacts between nanotubes has welded together to become CNT junctions. In the multiscale approach, a macroscopic analytical model is proposed to predict the overall thermal conductivity of the network, where model parameters such as intra-tube thermal resistance and inter-tube thermal conductance of CNT junctions are obtained from molecular dynamics (MD) simulations. Both the inter-tube thermal conductance and the intra-tube thermal resistance for the X- and EX-junctions are investigated, and their relationship with the interconnected bonds number of the junctions is discussed. The inter-tube junction thermal conductance is found to be two orders magnitude larger than contact conductance between two CNTs in close contact while intra-tube junction thermal resistance also increases significantly after sintering. Employing MD simulation results, the macroscopic model predicts that the intra-tube junction thermal resistance has a dominant effect on the overall network thermal conductivity. In addition, thermal conductivity of sintered network is found to decrease with increase in the fraction of welded contacts occurred in the network when the mass density or the tube length is large. However, thermal conductivity of the network will first increase and then decrease with the fraction of welded contacts in the network when both the mass density and the tube length are small. The network thermal conductivity versus mass density rho satisfies k(network) proportional to rho(gamma), the exponential gamma decrease with the increase of the intratube junction thermal resistance and the fraction of welded contacts. The results presented in this work will potentially have important implications in the design of CNT random networks for thermal management and thermoelectric applications. (C) 2013 Elsevier Ltd. All rights reserved.