In estimating flux at a point in a Monte Carlo calculation one estimator uses the uncollided flux at a detector from each sampled collision point. This method is shown to have infinite variance. The average value converges to the expected value but the error decreases asymptotically as the inverse cube root of the number of histories. By using the once collided flux and by proper choice of the intermediate collision point the variance may be made finite. Results of numerical experiments show the finite variance methods to be preferable.