While DNA detection using capillary electrophoresis has enabled improvements in both resolution and throughput, the use of CE – particularly with multiple dye channels – can introduce artifacts that can complicate analyses. Undetected pull‐up artifacts can pose a challenge to investigators, especially in low‐level samples, while partial pull‐up peaks can distort peak height balance within a locus and impact the downstream likelihood ratio. Current methods for addressing pull‐up are typically manually implemented. This study presents an effective alternative: a series of mathematical models, created using symbolic regression achieved through genetic programming. The models estimate the amount of pull‐up expected in a peak from a true allele for a given dye‐dye relationship and instrument type. This leads to the removal of artifactual pull‐up peaks and peak height corrections when pull‐up is present within true alleles. When models are used in conjunction with a dynamic threshold, pull‐up peaks were automatically detected and removed with an accuracy rate of 96.1%. The removal of partial pull‐up from true allele peaks led to a more accurate heterozygote balance for the affected locus. These models have been optimized for use with any analytical threshold and can be implemented by any lab using a 3100 or 3500 instrument series.