Actually, it is quite possible that Barfield himself wrote an excellent review, as part of is Preface to the book's second edition (1951 - the original being from 1928):
...[T]his book grew out of two empirical observations, first, that poetry reacts on the meanings of the words that it employs, and, secondly, that there appear to be two sorts of poetry... Thus, it claims to present, not merely a theory of poetic diction, but a theory of poetry: and not merely a theory of poetry, but a theory of knowledge.
Yes indeed, Barfield attempts all that, nor am I ambitious enough to summarize the theory of poetry that he takes 225 pages to unfold. I can, however, mention some of its salient features.
First, that he distinguishes between poetry and verse. Poetry, for Barfield, is a quality of language: poetic diction, as opposed to prosaic diction, either of which may appear in verse or prose form.
Poetic language, poetry, he suggests, is language that creates new meaning for (at least some of) the words it uses.
When words are selected and arranged in such a way that their meaning either arouses, or is obviously intended to arouse, aesthetic imagination, the result may be described as poetic diction... Meaning includes the whole content of a word, or of a group of words arranged in a particular order, other than the actual sounds of which they are composed.Indeed, Barfield eschews discussion of things like rhyme, alliteration, and so on, except passim in service of other birds he is hunting.
A great deal of the book is given over to a proposed revision of the commonly-held theories (as of 1928) of how language arises, and of how words "create" meaning in the mind of the reader/writer/speaker/hearer. He does not explicitly rejects, but thoroughly criticizes, the philological assumption that all modern languages are grown from simple linguistic "roots," suggesting that out grancestors actually had much more capacity for abstraction in language than we give them credit for - and, at the same time, much more capacity for concrete specificity.
He suggests that words that seemed to have two meanings - such as the Latin verb "ruo," which is translated (depending on context) either as "rush" or "fall." It may be, he says, that to the Latin mind, there was a single meaning -- a single concept that has relationship to both, but is not either of, the modern concepts "rush" and "fall." He traces one of the word's English descendants, ruin, through a series of meaning-concepts it has represented over the centuries.
I can't really grok his conclusions very well; as I said, I shall have to reread this book at some point. But it has something to do with the idea that poetic language occurs precisely when the speaker/writer says/writes something that the hearer/reader finds "strange" -- for a variety of meanings of "strange," which however basically come to "the hearer/reader is given a new way of seeing something."