Can we decode a person’s beliefs from measurements of their neural activity? In previous work, I disambiguated this question and offered a qualified “yes.” In this talk, I want to expand on two of the necessary qualifications. The first concerns the contrast between strict decoding, on the one hand, and what we might call “reconstruction,” on the other. I suggest that an important sense of the term “decoding” refers to a process in which non-neural sources of information are kept to a minimum. Once we help ourselves to sources of non-neural information, our capacity to reconstruct a person’s beliefs improves drastically, but only because it has more in common with our common practices of belief attribution than meets the eye. One goal of this part of the talk is to draw attention to the manner in which some contemporary paradigms for decoding mental content from fMRI data already, and in rather subtle ways, incorporate non-neural sources of information. The second qualification concerns the strength of the neural evidence. Here, I ask whether it is possible that behavioral evidence for the claim that S believes that P could be overturned by purely neural evidence. The answer to this question depends on philosophical theories about the nature of belief. I show that a popular combination of philosophical views about the nature belief entails that the answer is “no.”