In a predictable natural selection process, herbicides select for adaptive alleles that allow weed populations to survive. These resistance alleles may be available immediately from the standing genetic variation within the population or may arise from immigration via pollen or seeds from other populations. Moreover, because all populations are constantly generating new mutant genotypes by de novo mutations, resistant mutants may arise spontaneously in any herbicide-sensitive weed population. Recognizing that the relative contribution of each of these three sources of resistance alleles influences what strategies should be applied to counteract herbicide-resistance evolution, we aimed to add experimental information to the resistance evolutionary framework. Specifically, the objectives of this experiment were to determine the de novo mutation rate conferring herbicide resistance in a natural plant population and to test the hypothesis that the mutation rate increases when plants are stressed by sublethal herbicide exposure. We used grain amaranth (Amaranthus hypochondriacus L.) and resistance to acetolactate synthase (ALS)-inhibiting herbicides as a model system to discover spontaneous herbicide-resistant mutants. After screening 70.8 million plants, however, we detected no spontaneous resistant genotypes, indicating the probability of finding a spontaneous ALS-resistant mutant in a given sensitive population is lower than 1.4 × 10−8. This empirically determined upper limit is lower than expected from theoretical calculations based on previous studies. We found no evidence that herbicide stress increased the mutation rate, but were not able to robustly test this hypothesis. The results found in this study indicate that de novo mutations conferring herbicide resistance might occur at lower frequencies than previously expected.