This study investigates the false-positive rates of fMRI analyses using parametric and nonparametric methods. The researchers used real resting-state fMRI data and 3 million random task group analyses to evaluate the statistical software packages SPM, FSL, and AFNI, as well as a nonparametric permutation method. They found that while parametric methods were conservative for voxelwise inference, they were invalid for clusterwise inference, leading to high false-positive rates (up to 70%) compared to the nominal 5% rate. In contrast, the nonparametric permutation test produced nominal results for both voxelwise and clusterwise inference. The study highlights the need for validating statistical methods in neuroimaging and suggests that spatial autocorrelation functions not following a Gaussian shape are a primary cause of invalid cluster inferences. The findings also indicate that the use of nonstationary smoothness assumptions in parametric methods can further contribute to inflated false positives.This study investigates the false-positive rates of fMRI analyses using parametric and nonparametric methods. The researchers used real resting-state fMRI data and 3 million random task group analyses to evaluate the statistical software packages SPM, FSL, and AFNI, as well as a nonparametric permutation method. They found that while parametric methods were conservative for voxelwise inference, they were invalid for clusterwise inference, leading to high false-positive rates (up to 70%) compared to the nominal 5% rate. In contrast, the nonparametric permutation test produced nominal results for both voxelwise and clusterwise inference. The study highlights the need for validating statistical methods in neuroimaging and suggests that spatial autocorrelation functions not following a Gaussian shape are a primary cause of invalid cluster inferences. The findings also indicate that the use of nonstationary smoothness assumptions in parametric methods can further contribute to inflated false positives.