Abstract. We study the structure of infinite-to-one continuous codes between subshifts of finite type and the behaviour of Markov measures under such codes. We show that if an infinite-to-one code lifts one Markov measure to a Markov measure, then it lifts each Markov measure to uncountably many Markov measures and the fibre over each Markov measure is isomorphic to any other fibre. Calling such a code Markovian, we characterize Markovian codes through pressure. We show that a simple condition on periodic points, necessary for the existence of a code between two subshifts of finite type, is sufficient to construct a Markovian code. Several classes of Markovian codes are studied in the process of proving, illustrating and providing contrast to the main results. A number of examples and counterexamples are given; in particular, we give a continuous code between two Bernoulli shifts such that the defining vector of the image is not a clustering of the defining vector of the domain.0. Introduction. This paper is concerned with the structure of infinite-to-one continuous codes between subshifts of finite type and with the behaviour of Markov measures under such codes. The emergence of notable interest in infinite-to-one codes is recent. This interest was partly fueled by the development of Krieger's marker method [11, 12] and is reflected by the papers [3 and 15]. Previous research had concentrated on bounded-to-one codes and yielded considerable information about these. We list [1,4,[7][8][9][10][16][17][18][19][20] for examples of material dealing with bounded-to-one codes. The properties of bounded-to-one codes constituted one of our main motivations; it may be useful to compare our results with these.After a section on definitions and notation, we will give three examples. The first example concerns infinite-to-one analogues of the resolving codes used in [1, 16 and 17]. The main point about these is that they lift each Markov measure to uncountably many Markov measures. In [15], Marcus, Petersen and Williams construct infinite-to-one codes through which no Markov measure lifts to a Markov measure. Recall that, through a bounded-to-one code, each Markov measure lifts to a unique Markov measure [20]. This lifting process does not alter the memory of the measure when the code is 1-block. However, in the infinite-to-one case there may be strict increases in memory not warranted by the block length of the code: our third (peculiar memory) example is a 1-block code such that no 1-step Markov measure lifts to a 1-step Markov measure, while each 1-step Markov measure has uncountably many 2-step preimages. In addition, this code sends every 1-step Markov