Experiment 1 reanalyzed data from a counterexample study and found that including spatial information and nonlinear classifiers provided above-chance accuracy in decoding tasks from eye movements. Experiment 2 applied the counterexample's unbiased procedure to a new dataset and tasks, successfully decoding tasks above chance. This supports Yarbus's hypothesis that eye movements reveal cognitive tasks. However, task decoding is not guaranteed - it depends on factors like task separability, stimulus information, and observer knowledge. The findings help define guidelines for successful versus unsuccessful task decoding from eye gaze.