Autoclaved aerated concrete (AAC) has widely been utilized as a lightweight, porous insulation material for energy-efficient buildings. The knowledge on the thermal conductivity of AAC is required for thermal design of building envelopes. The effective thermal conductivity of AAC is strongly dependent on the moisture content. Such dependence, however, is not well documented in available literature. In this work, AAC bricks with three different bulk densities of 415, 520, and 630 kg/m3, were obtained as the raw materials, and the samples were prepared by humidification to a set of moisture content levels up to 100% by mass. The effective thermal conductivity of the moisturized samples was measured by means of the transient plane source technique. Meanwhile, fractal models for predicting the effective thermal conductivity were proposed based on construction of the porous structure of AAC by self-similar Sierpinsld carpet. A two-phase fractal model was first proposed for dry AAC samples, and then an extension to a three-phase model was developed by considering the presence of water phase in the pores for unsaturated, moist samples. It was shown that the thermal conductivity increases with increasing the moisture content, by a factor up to 3.8 over the studied range of moisture content, following a two-section piecewise linear variation. A high-to-low slope change was found to be around a moisture content of 15% for all the AAC samples. A correlation was proposed for the measured thermal conductivity as a function of both moisture content and porosity. Appropriate parameters for the two-phase model were determined by comparing the predicted results to the measured data at dry state. The three-phase fractal model was exhibited to be able to predict the hygric dependence of thermal conductivity. The discrepancy among the predictions by the three-phase model with different geometric parameters was discussed in relation to the constructed pore structures. The predicted results by the two configurations of the three-phase model, i.e., with and without considering the presence of connected water bridges in the pores, were also presented. A reasonable elimination of the presence of connected bridges was shown to lead to better predictions in the low moisture content regime. (C) 2015 Elsevier Ltd. All rights reserved.