Although the roles of symbolic numerical magnitude processing (SNMP) and working memory (WM) as predictors of mathematics performance are well acknowledged, studies examining their joint effects are still scarce. In this study, we investigated whether we can identify different performance profiles of SNMP (i.e., comparison of 1- and 2-digit numbers) and WM (i.e., verbal, visuospatial, and central executive) among Norwegian first-graders (N = 256, Mage = 6 y. 9 m.), and how those profiles predict performance in counting and arithmetic, while controlling for several demographics and cognitive skills. Using latent class cluster analysis, four groups were identified: 1) weak SNMP (33.6%), 2) strong SNMP (25.8%), 3) weak SNMP and WM (double weakness) (23.4%), and 4) strong WM (17.2%). The group differences in counting and arithmetic were significant with explained variance ranging from 21% to 30%, and remained so even after taking into account the effects of age, gender, parents’ educational level, status of second language learner, nonverbal intelligence, rapid automatized naming, and word comprehension. Our findings suggest that children may display relative strengths in SNMP and WM skills, and that they both have a unique, even compensatory role in children's mathematics performance. Yet, their individual contribution might depend on the types of skill in question (e.g., arithmetic fluency vs. word problem solving). Particularly SNMP skills might serve as a marker for future difficulties in learning mathematics.