Since all faces share the same first-order configuration, individual faces are recognized by subtle differences in their individual features and second-order configuration. This study asked two questions about the ERP signatures of featural and configural face processing. First, which ERP components are sensitive to subtle modifications of features and second-order configuration? Secondly, does explicitly instructing participants to attend to either the features or the configuration of face stimuli influence the neural processes elicited by these stimuli? In the first part of this experiment, pairs of faces with differences in their features, in their second-order configuration or no difference were presented, with participants making same/different judgments. In the second part, participants were asked to focus on either the features or configuration in pairs of faces and to judge if there was any difference in that aspect while ignoring other differences. The P1 and N170 ERP components were not influenced by feature or second-order configuration differences in faces, indicating that these components could be more sensitive to manipulations that disrupt the first-order configuration of a face. In contrast, the P2 was larger in amplitude for faces with a configuration modification than for faces with a feature modification or for the original face. One explanation of these results is that processing the configuration of a face elicits more visual cortical feedback than processing the features. Finally, modulation of the P300 confirmed that the tasks presented in Part 2 were successful in varying the attention focus on the features or configuration.