The validity of body mass index (BMI) specifiers for anorexia nervosa (AN) has been questioned, but their applicability to inpatients with extremely low BMIs and their prognostic validity are currently unknown. Therefore, we designed this study: (a) to test current BMI specifiers in severe inpatients; (b) to explore a "very extreme" specifier (VE-AN; BMI ≤ 13.5); and (c) to verify inpatients' hospitalization outcome according to BMI severity. We enrolled 168 inpatients with AN completing the following: Eating disorder Examination-Questionnaire, Eating Disorder Inventory-2, State-Trait Anxiety Inventory, Beck Depression Inventory, Body Shape Questionnaire, and EQ-5D-VAS. According to the current BMI classification, those with a BMI < 15 versus those with non-extreme AN (NE-AN, BMI ≥ 15) differed on all measures but the quality of life with those with NE-AN reporting more impaired scores on all measures. Adopting an exploratory classification comparing VE-AN, extreme AN (E-AN, BMI = 13.6-14.99), and NE-AN, no differences emerged between VE-AN and E-AN, while those with NE-AN reported significantly more impaired scores on all variables while the quality of life again did not differ across groups. Hospitalization outcome improved for all groups, independently of BMI. Groups differed concerning the length of stay that mirrored BMI severity and impacted also hospitalization outcomes. Taken together, our data support the lack of validity of current BMI specifiers in AN, even in the acute setting. Moreover, the exploratory subgroup of patients with BMI ≤ 13.5 did not delineate a clinically different group.