The wait for ASD evaluation dramatically increases with age, with wait times of a year or more common as children reach preschool. Even when appointments become available, families from traditionally underserved groups struggle to access care. Addressing care disparities requires designing identification tools and processes specifically for and with individuals most at-risk for health inequities. This work describes the development of a novel telemedicine-based ASD assessment tool, the TELE-ASD-PEDS--Preschool (TAP-Preschool). We applied machine learning models to a clinical data set of preschoolers with ASD and other developmental concerns (n = 914) to generate behavioral targets that best distinguish ASD and non-ASD features. We conducted focus groups with clinicians, early interventionists, and parents of children with ASD from traditionally underrepresented racial/ethnic and linguistic groups. Focus group themes and machine learning analyses were used to generate a play-based instrument with assessment tasks and scoring procedures based on the child's language (i.e., TAP-P Verbal, TAP-P Non-verbal). TAP-P procedures were piloted with 30 families. Use of the instrument in isolation (i.e., without history or collateral information) yielded accurate diagnostic classification in 63% of cases. Children with existing ASD diagnoses received higher TAP-P scores, relative to children with other developmental concerns. Clinician diagnostic accuracy and certainty were higher when confirming existing ASD diagnoses (80% agreement) than when ruling out ASD in children with other developmental concerns (30% agreement). Utilizing an equity approach to understand the functionality and impact of tele-assessment for preschool children has potential to transform the ASD evaluation process and improve care access.