Finite element models are greatly relied on by the packaging community to predict stresses that develop within microelectronic packages. While of great value, model predictions do not always accurately predict or explain package failures. Particularly, there has been much discussion about which models best describe the behavior of polymeric composites such as the underfill. Additionally, current models do not account for material heterogeneity and non-perfect geometries which are conditions commonly seen in real packages. How such imperfections impact local stresses that often drive failure is unknown. We therefore present a method of directly measuring local stresses in the underfill region using carbon nanotubes as sensors and show how this method can be used to improve finite element models and to assess the impact of conditions that are difficult to model. © 2010 IEEE.