Background <p>Two-level osteotomy has emerged as an effective technique for addressing severe kyphosis secondary to ankylosing spondylitis (AS). Despite its efficacy, there remains a lack of consensus regarding the criteria for determining the necessity of two-level osteotomy. This study aimed to investigate precise and direct preoperative predictors for selection of two-level osteotomy in patients with severe AS kyphosis.</p> Methods <p>A retrospective cohort of 101 AS patients was analyzed, including 33 patients who underwent two-level modified pedicle subtraction osteotomy (PSO) and 68 patients who underwent one-level modified PSO. Radiographic parameters, including pelvic tilt, pelvic incidence (PI), sacral slope (SS), lumbar lordosis (LL), thoracic kyphosis (TK), global kyphosis (GK), sagittal vertical axis (SVA), chin-brow vertical angle (CBVA), and osteotomized vertebral angle, were measured. Clinical outcomes were assessed using Oswestry Disability Index and Scoliosis Research Society-22 questionnaire. Comparative analyses of radiographic and clinical outcomes were conducted across different patient groups. Preoperative predictors for selecting two-level osteotomy were identified through receiver-operating characteristic curve analysis and logistic regression analysis.</p> Results <p>Patients undergoing two-level osteotomy exhibited significantly higher preoperative parameters of SS, LL, TK, GK, SVA, and CBVA compared to those receiving one-level osteotomy (<i>P</i> &lt; 0.05). Notably, CBVA, GK, and SVA were identified as the most influential parameters influencing the selection of two-level osteotomy, with optimal threshold values of 63.9°, 91.1°, and 25.4&#xa0;cm, respectively. Logistic regression analysis revealed CBVA and GK as independent predictors for selecting two-level osteotomy (<i>P</i> &lt; 0.05). Patients undergoing lumbar region two-level osteotomy demonstrated larger SVA (<i>P</i> &lt; 0.05) and pelvic incidence (PI; <i>P</i> = 0.267), whereas those with thoracic and lumbar osteotomy exhibited increased TK (<i>P</i> = 0.465). All patients achieved favorable clinical outcomes at final follow-up (<i>P</i> &lt; 0.05).</p> Conclusions <p>Preoperative CBVA, GK, and SVA are key parameters influencing the selection of two-level modified PSO for AS kyphosis. Specifically, preoperative CBVA &gt; 63.9° and GK &gt; 91.1° serve as independent predictors, with SVA &gt; 25.4&#xa0;cm acting as an auxiliary criterion. The choice of osteotomy sites is predominantly influenced by preoperative SVA, PI, and TK measurements.</p> Level of evidence <p>Level IV, therapeutic study.</p>