Thickness loss caused by static and dynamic loads is one of the most important quality factors in carpets. This property is influenced by pile yarn characteristics and carpet construction parameters. Exploring the relationship between thickness loss and effective factors is highly significant to optimize the selection of the variables. Soft computing approaches, which are potent data-modeling tools in capturing complex input-output relationships, seem to be the powerful technique to decipher the above-mentioned relationship. This paper presents two different modeling methodologies for predicting thickness loss of Persian hand-knotted carpets. At first, several topologies with different architectures were used to get the best neural-network model. Since, ANN model is a black box and did not succeed in indicating inter-relationship between input and output parameters, gene expression programming (GEP) is presented here as another intelligent algorithm to predict thickness loss of the carpets. Our study showed that, GEP model has a significant priority over the ANN. The correlation coefficient (R-value) and mean square error (MSE) for GEP model were 0.950 and 0.219 respectively, while these parameters were 0.707 and 0.510 for ANN model. This indicates the desirable predictive power of GEP algorithm. Based on the proposed models, the dominant parameter on thickness loss found to be knot density and content of slipe wool.