Suppose we observe many emeralds which are all green. This observation usually provides good evidence that all emeralds are green. However, the emeralds we have observed are also all grue: either green and already observed or blue and not yet observed. We usually do not think that our observation provides good evidence that all emeralds are grue. Why not? This is Goodman’s ‘New Riddle of Induction’. I argue that if we are in the best case for inductive reasoning, we have reason to assign a very low probability to all emeralds being grue before seeing any evidence. My argument appeals to random sampling and the observation-independence of green.