PURPOSE: Random-start controlled ovarian hyperstimulation (COH) is an established strategy for fertility preservation (FP) in cancer patients. Although previous studies have demonstrated its efficacy using a binary classification of menstrual cycle phases (follicular vs. luteal), this approach might not fully reflect dynamic endocrine changes throughout the cycle. This study used a subdivided phase classification to evaluate the ovarian response according to the menstrual cycle phase at random-start COH initiation and further assessed the outcomes according to the ovarian reserve. METHODS: This retrospective study included 255 cancer patients who underwent random-start COH. Patients were categorized into five menstrual cycle phases at the time of stimulation initiation: early follicular (EF), late follicular (LF), early luteal (EL), mid-luteal (ML), and late luteal (LL). Oocyte yield, total gonadotropin dose, and stimulation duration were compared among groups. A subgroup analysis was conducted in patients with low ovarian reserve (anti-Müllerian hormone [AMH] < 1.2 ng/mL). RESULTS: Total oocyte yield was significantly higher in the LL group than in the EF group, though with a greater total gonadotropin dose and longer stimulation duration. Among patients with low ovarian reserve, luteal phase initiation was associated with significantly higher total and mature metaphase II (MII) oocyte counts compared with follicular phase initiation. CONCLUSIONS: The menstrual cycle phase at stimulation initiation influences ovarian response during random-start COH. The LL phase represents an effective window for stimulation initiation, and luteal phase initiation is particularly advantageous for maximizing oocyte yield in patients with low ovarian reserve who require urgent FP.