In this thesis propose a solution for spelling correction in Spanish based in BERT. It starts from the collecting and processing of Spanish corpus, a explanation of generation of error based on linguistic research about spelling errors in Spanish. After expose how BERT pre-trained as Masked Language Model (MLM) was used to correct tagged errors and propose an approach to make a detection network for adapt it for real spelling correction task. Finally show the experiments carried out in CNN-CHAR-MODEL, and list the expected experiments in proposed model for spanish misspelling correction with BERT.