The main objective of this study was to estimate the selection accuracy and to predict the genetic gain in cassava breeding using genomic selection methodologies. We evaluated 358 cassava genotypes for the following traits: shoot weight (SW), fresh root yield (FRY), starch fraction amylose content (AC), dry matter content (DMC), and starch yield (S-Y). Genotyping was performed using 390 single nucleotide polymorphisms (SNPs), which were used as covariates in the random regression-best linear unbiased prediction model for genomic selection. The heritability values detected by markers for the SW, FRY, AC, DMC, and S-Y traits were 0.25, 0.25, 0.03, 0.20, and 0.26, respectively. Because the low heritability detected for AC, this trait was eliminated from further analysis. Using only the most informative SNPs (118, 92, 56, and 97 SNPs for SW, FRY, DMC, and S-Y, respectively) we observed higher selection accuracy which were 0.83, 0.76, 0.67, and 0.77, respectively to SW, FRY, DMC, and S-Y. With these levels of accuracy and considering a selection cycle reduced by half the time, the theoretical gains with genomic selection compared to phenotypic selection for DMC, FRY, and SW would be 39.42 %, 56.90 %, and 73.96 %, respectively. These results indicate that in the cassava, genomic selection can substantially speed up selection cycles, thereby increasing gains per unit time. Although there are high expectations for incorporating this strategy into breeding programs, we still need to validate the model for other traits and evaluate whether the selection accuracy can be improved using more SNPs.