Lead (Pb2+) which is one of the most important heavy metals found in water sources, has a toxic effect on living things in aquatic environment. Therefore, the removal of Pb2+ ions from wastewater is very important if its concentration is above the determined discharge limits. Due to its advantage such as high efficiency, the adsorption process is a widely used successful technique for heavy metals removal from an aqueous solutions. On the other hand, determining the adsorption efficiencies of different adsorbents experimentally is both costly, and time-consuming considering that there are large number of process variables. Therefore, ANN can be used to make theoretical predictions with high efficiency in this treatment process. In this study, modeling of Pb2+ removal from an aqueous solution by using biochars that produced by pyrolyzing of hazelnut, and walnut shell was studied at different adsorption conditions; pH (2.5-5), temperature (25-45oC), initial Pb2+ concentration (15-45 mg/L), adsorbent amount (1-3 g/L), and mixing speed (200-600 rpm). The purpose of modeling studies with ANN approach was to estimate lead ions removal (%) as an output. Inputs for ANN modeling approach were selected as pH, initial Pb2+ concentration, temperature, adsorbent dosage, and mixing speed. Experimental data were categorized 50:25:25 for two adsorption systems. Levenberg-Marquardt (LM) was preferred as a training function, and tansig was used as an activation function. The number of hidden neurons in the hidden layer was found by trial, and error. Values of correlation coefficient (R2), and Mean Square Error (MSE) were taken to be performance criteria of the ANN modeling. R2 values were found to be 97%, and 98% for biochars derived from walnut, and hazelnut shells, respectively. Results showed that ANN is an effective tool for modeling of adsorption system for the removal of lead ions from an aqueous solution. Additionally, different training algorithms such as Bayesian Regularization (BR), and Scaled Conjugate Gradient (SCG) were used to compare the prediction capability.