The present study deals with the first impression about a person’s intelligence, based solely on their text. We attempted to calculate the accuracy of a reader’s judgment, and to determine which features of the text are associated with both the reader’s impression as well as the author’s intelligence test scores (Brunswik lens model). The study involved 116 participants aged 16 to 23 years. Each of them read and evaluated 4 out of 20 texts that were written by other participants in an earlier study. Participants evaluated both the general impression of the author’s intelligence and of the text itself, scoring responses to questions such as “How smart is the author of this text?” and to other text features that may be associated with intelligence judgement, including “vocabulary richness” and “grammar complexity”. In addition, objective indicators such as readability (Flesch-Kincaid score), length, and relative errors were calculated for the texts. The accuracy of reader judgments, estimated in a text-based analysis as the Spearman correlation coefficient between the “how smart is the author” score and the actual author test scores, did not reach statistical significance. However, the data indicate the presence of a nonlinear relationship. As revealed by a participant-based analysis, authors from the lower quartile (based on test scores) tend to be rated lower by participants as well. Readability, vocabulary richness, grammar complexity, and the relative number of words longer than six letters all were utilized by the reader participants in their judgments (i.e., correlated with impression scores) and at the same time were valid (i.e., correlated with test scores). We discuss the problem of the narrowness of scientific and psychological impressions on intelligence, compared to everyday perceptions