The commonly used grade-level discrepancy assessment, which relies on achievement test scores to identify children with mathematics learning disabilities (MLD or MD), can be biased and its effectiveness varies across children’s cognitive levels, educational backgrounds of families, school contexts, and children’s learning styles. As a consequence, the identification of children with MD often lacks accuracy. With an integration of interventions and interactive evaluation, dynamic assessment is developed to identify children’s potential for learning which is less influenced by their family and school backgrounds mentioned above. Unlike traditional achievement tests, dynamic assessment applies new criterion on MD that focus on children’s potential for learning mathematics and can reduce above-mentioned unfavorable effect to some extent. The current study attempted to demonstrate that the addition of dynamic assessment to traditional achievement tests can help identify MD children and the subgroups of MD children with increased accuracy and objectivity. Dynamic assessment has a unique effect on rapid and accurate identifying children with mathematics learning disabilities. Following the standard orientation model proposed by Proctor and Prevatt (2003), the treatment group included 30 fourth-grade children with mathematics learning disabilities, while the control group consisted of 30 children who were matched on Raven intelligence test scores. Four subtests (Rhyming Words, Auditory Digit Sequence, Visual Matrix and Mapping and Directions) of the “Swanson Cognitive Processing Test (S-CPT)” were administrated to assess children’s working memory. The assessment included initial score, gain score, maintain score, difference score, stable score, guide score and strategy score, collected in four phases: pretest, intervention, posttest, and delayed posttest. The differences in work memory task scores before and after the dynamic interventions, were used to categorize the 30 children in the treatment group into different subgroups. The results of factor analyses revealed two factors (original cognitive abilities and potential cognitive abilities), with the factor of potential cognitive abilities explained additional 19% of variance in children’s mathematics achievement. By integrating the factor of potential cognitive abilities as a core criterion, the initial group of 30 MD children can be further categorized into two subgroups: insufficient development group (n = 11, 37%) and developing deficit group (n = 19, 63%). The insufficient development subgroup differed from the developing deficit subgroup in several ways, including higher gain score, maintain score, stable score, difference score and guide score scores that were comparable to normal children and improved mathematics achievement test scores after one year. These findings point to the benefit of using dynamic assessment to identify subgroups of MD children, in comparison to traditional achievement tests which failed to distinguish the insufficient development subgroup from the developing deficit subgroup. In conclusion, by integrating interventions and interactive evaluations, dynamic assessment provides a unique way to explore children’s potential for learning that cannot be captured by traditional achievement tests. In addition, dynamic assessment complements traditional achievement tests by identifying subgroups of MD children: the insufficient development subgroup and the developing deficit subgroup; the first subgroups may otherwise be mis-labeled as learning disable children when using traditional achievement tests alone. The identification of subgroups of MD children can also lead to diversified intervention effort that may be more effective for one subgroup but not the other.