We provide new evidence about language assimilation and its effect on test scores using data from two rounds (conducted approximately six years apart) of the New Immigrants Survey (NIS). As part of the NIS interviews, U.S. born and foreign-born children of immigrants were asked to take Woodcock-Johnson achievement test. In both rounds, prior to the administration of tests, children of Hispanic origin were randomly assigned to take the tests either in Spanish or in English. Therefore, we can attribute the difference in scores to language proficiency and directly estimate the rate of assimilation. Our results suggest that in reading tests, U.S. born children of Hispanic immigrants perform better, when they are assigned to take the tests in English, and the advantage remains stable across two rounds of interviews. However, there is substantial heterogeneity. For example, U.S. born children at the top of score distribution perform better when they take tests in Spanish. Foreign-born children of Hispanic immigrants exhibit Spanish dominance during the first round, but it declines and in some cases completely disappears by the second round. We find that foreign-born children who immigrated to the U.S. after age six, exhibit Spanish dominance in reading tests in the first round. However, during the six years between interviews, Spanish dominance disappears among foreign-born children who immigrated between the ages of six and eight (in reading) and in all children (in math). Moreover, for children who still have Spanish dominance in reading, the score differences have narrowed.