Latent models assert that a non-observed concept may explain the differences among observed scores. Therefore, if the observed scores are known, the latent scores can be predicted. However, other latent dimensions might have an effect on the observed scores, thereafter, the estimation of the latent scores is biased. In this context, intervenient variables, such as depression and anxiety, tend to decrease the subject’s ability to properly answer the items of an intelligence test, even if the intelligence is preserved. The study aims to develop a model, based on SEM, to estimate the intelligence scores freely from the depression and anxiety bias. Method: Brazilian undergraduate students (N = 168) volunteer participated in this study. The students responded the following measures: i- Abstract and Reasoning Intelligence Test; ii- Baptista’s Depression Inventory; iii- Questionnaire of Anxiety Related to Achievement. The data analysis was performed on two steps: First, we test a model with direct paths between anxiety-intelligence and depression-intelligence; second, we test direct paths between anxiety, depression, and the items (observed) of the intelligence test. Results: The first model indicated a significant direct effect from the anxiety to the intelligence (latent); meanwhile the effect from the depression to the intelligence was not significant, and we decide to exclude this variable from the upcoming models. In the second model, the anxiety had a direct effect on two items of the intelligence test (standard effects = .29 and .30). Then, the last model showed that only less than 10% of the items were affected by the anxiety, after controlling the intelligence latent variance. In other words, the anxiety did not shrink the odds to correct answer the majority of the items of the intelligence test; consequentially, the estimation of the intelligence latent scores were not affect by the anxiety as well