But in fact the criticism commonly made is not that, happening as described, visions and hallucinations happened by chance; but that they did not happen as described. This objection deserves careful consideration. It must, I think, be admitted that a proportion, perhaps a large proportion, even of the cases obtained at first-hand are so far inaccurate as to have comparatively small value for scientific purposes; and of the residue, in which the central fact of an unusual subjective experience on the part of the percipient and its coincidence with some external event is fairly well established, it is possible that the details are frequently--and where the record is not made until some years after the event, generally--untrustworthy. In order to estimate the nature and probable extent of these defects, it is proposed briefly to pass in review the various kinds of error to which testimony is liable, and to note their bearings on the question at issue. _Errors of Observation._ Errors of observation are here of very little importance. The thing to be observed is, of course, the percipient's own sensations. In subsequent conversation he may exaggerate the exceptional nature of the impression; but he can hardly make a mistake at the time in observing what is purely subjective. If a man calls green what we call red, we may conclude that he is colour-blind; and if he asserts that he sees a human figure where we see none, that he is hallucinated; but in neither case have we warrant for saying that he is making an erroneous statement about his own sensations. _Errors of Inference._ But his interpretation of what he sees is a different matter. Not indeed that the mistake commonly made of taking a hallucination at the time for a figure of flesh and blood, and subsequently for a hypothetical entity of another kind, directly affects the percipient's testimony. So long as the witness accurately describes what he saw, it matters little whether he believes in telepathic hallucinations, or in black magic, ghosts, or the Himalayan Brothers. But there are one or two errors of inference of sufficient importance to deserve notice.