Kernel density estimation (KDE) has been prominently used to measure poverty from grouped data (representing mean incomes of a small number of population quantiles). In thispaper I analyze the performance of this method. Using Monte Carlo simulations for plausible theoretical distributions and unit data from several household surveys, I compare KDE-based poverty estimates with their true and survey counterparts. It is shown that the technique gives rise to biases in poverty whose sign and magnitude vary with the smoothing parameter, the kernel, the number of data-points analyzed, and the poverty indicators used. I also demonstratethat KDE-based global poverty rates and headcounts are highly sensitive to the choice ofsmoothing parameter. Depending on the parameter, the estimated proportion of ‘$1/day poor’ in 2000 varies by a factor of 1.8, while the estimated number of ‘$2/day poor’ in 2000 varies by 287 million people. These findings give rise to concern about the validity and robustness of kernel density estimation in poverty analysis. However, they provide a framework for interpretation of existing results using this technique.