In the area of researches about effective schools, this study investigates specially the teacher effect in the learning of reading by students of the second and third grades of elementary school. Our interest is to investigate which practices are related to the increase of learning of the different skills that need to be constructed during the learning of a native language, and also to check if the impacts of these practices are different in each grade of elementary school. The analyzed data were collected in a longitudinal project called GERES – Estudo Longitudinal da Geração Escolar 2005. The subsample uses cognitive data of 17,628 students, distributed in 671 groups of 225 schools. The research took place in county, state, federal, and private schools of the following cities: Rio de Janeiro (RJ), Belo Horizonte (MG), Campinas (SP), and Campo Grande (MS). The contextual data of the subsample correspond to information provided by the teachers of those groups. This study may be condensed in three complementary investigations. The first one corresponds to the pedagogical interpretation of the proficiency scale, which is mandatory to the comprehension of any posterior result, considering that the numerical values of the sample do not have intrinsic meaning. Therefore, tests items were examined in detail, according to their distribution in the scale’s proficiency ranges, in order to make clear what the students are able to do in each item. After this, a new pedagogical analysis was done with the intention of making the group of skills coherent, interpreting their different learning levels. In the second investigation, we search for evidences of the existence of subscales of skills that are internally related to the learning of reading. At the same time, we investigate if the practices declared by the teachers may also be expressed by scales. The constructs were initially defined by conceptual analysis and later by statistic procedures, which were necessary for their authentication. Finally, in the third investigation, which was the main objective of this study, we examine the effect of Portuguese language practices on the learning of the students who were evaluated along the research. The methodology we used for this association was based on the Hierarchical Multilevel Models (student and teacher levels). This methodology is ideal for researches based on hierarchically aligned data, what happens to educational research. The results of the research reveal good measures for the reading practices scales. The scales obtained were: (i) teacher reads for the students, (ii) teacher asks the students to read in silence, and (iii) teacher asks the students to read aloud. Although the results were not so statistically expressive, the scales we obtained for the reading skills were: (i) reading code processing; (ii) localizing explicit information; (iii) integrating text and context; and (iv) taking up discursive aspects from the text. The association between Portuguese language practices and proficiency has showed different achievements by the students. It varied mainly according to the skill, the school grade, and the teachers’ practices. Some of the results are that: (i) emphasizing traditional methods showed a negative relation with the learning of reading code processing and localizing explicit information in the second grade; (ii) in the third grade, highlighting copy, dictation, and calligraphy showed a negative relation to students’ learning in all the researched skills; and (iii) it was also observed that when the teacher reads for the students, in the third grade, it aggregates relevant knowledge to the students in all the researched skills, being this practice the only one that maintained significance in the research.