This paper studies certain aspects of language processing during reading: it describes the role of word length information obtained by parafoveal vision. Which properties of a word are processed parafoveally (that is, before the eyes actually move to it) and how this information is used are among the least understood questions in reading studies. The authors' goal was to test the length constraint hypothesis (LCH) according to which parafoveally obtained information about word length is used for word recognition, namely, to constrain the set of possible lexical candidates for the word to be recognized. Opponents of the LCH assume that this information is important only to determine where to direct our eyes during the next saccade: the preferred landing position for word processing is slightly to the left of the center. The results of the previous experiments testing this hypothesis are controversial. In this paper, the authors address several problems identified in earlier studies in an eye-tracking experiment on Russian. 24 adult Russian speakers participated in the experiment. The authors used the gaze-contingent boundary paradigm. This method involves eye-tracking while participants are reading sentences. Before the reader's eyes cross a particular point in front of the target word (so-called invisible boundary), another word or pseudoword (a preview) is shown on the screen. The preview is replaced by the target word while the eyes are moving across this boundary. Visual information processing is suppressed during saccades, so the actual change is not noticed, but its effects can be measured. There were 48 target sentences in three conditions: previews of the same length as targets, longer previews (in both cases, previews were orthographically similar to targets) and previews identical to targets. In the second condition, previews were only two letters longer than targets to minimize the difference between landing positions in this condition and two other ones. The data about eye movements were collected using an SR Eyelink 1000 plus eye tracker system. Statistical analysis using linear-mixed modeling revealed significant effects of length in the eyetracking measures reflecting early processing stages: the first fixation duration, single fixation duration and gaze duration were longer in the longer-preview condition than in the same-length condition. Most previous studies used pseudowords as previews, while the authors used real words, which could make the length effect more pronounced. To make sure that this result was not due to a non-optimal landing position (if the lengths of the preview and the target differ, the eyes may be directed to a non-optimal viewing position in the target), the authors included initial landing position to the model as a covariate. Hence, the authors concluded that their findings support the LCH: the readers obtain the information about word length parafoveally and use this information not only to plan the next eye movement, but also for word recognition.