The paradox of inference is based on the thesis that inferences cannot be both deductively valid and useful because an inference, in order to be useful, must advance us to a new conclusion, but a deductively valid inference cannot, on principle, do that, since its conclusions are contained in its premises. Falsification is a deductively valid inference as well. Its adherents should therefore respond to the paradox of inference. The paper elaborates on the view that critical, deductively valid arguments, e.g. the falsifying mode of inference modus tollens, do not advance those who use them to new knowledge, yet are useful because they help them to eliminate criticized theories. This view is confronted with the traditional response, according to which deductively valid inferences advance us to subjectively new conclusions and thus extend our subjective knowledge.