Spatial skills are associated with mathematics skills, but it is unclear if spatial training transfers to mathematics skills for preschoolers, especially from underserved communities. The current study tested (a) whether spatial training benefited preschoolers' spatial and mathematics skills, (b) if the type of feedback provided during spatial training differentially influenced children's spatial and mathematics skills, and (c) if the spatial training's effects varied by socioeconomic status (SES). Preschoolers (N = 187) were randomly assigned to either a 'business-as-usual' control or one of three spatial training groups (modeling and feedback [MF]; gesture feedback [GF]; spatial language feedback [SLF]). Three-year-olds were trained to construct puzzles to match a model composed of various geometric shapes. New models were created similar to the 2-dimensional trials of the Test of Spatial Assembly (TOSA). Training was given once per week for 5 weeks. Preschoolers were pretested and posttested on 2D and 3D TOSA trials, spatial vocabulary, shape identification, and 2 mathematics assessments. Results indicate that first, any spatial training improved preschoolers' 2D TOSA performance, although a significant interaction with SES indicated improvement was driven by low-SES children. Furthermore, low-SES children showed greatest gains on the 2D TOSA with MF and GF. Second, MF and GF improved low-SES children's performance on the 3D TOSA. Third, only low-SES children with MF saw improvements in far-transfer to mathematics (Woodcock-Johnson: Applied Problems, but not the Test of Early Mathematical Ability). Results indicate that, especially for low-income learners, spatial training can improve children's early spatial and mathematics skills.