Background: Mosunetuzumab is a CD20 × CD3 T-cell-engaging bispecific monoclonal antibody that redirects T cells to eliminate malignant B cells. In a phase 1 study, mosunetuzumab was well tolerated and active in patients with relapsed or refractory B-cell lymphoma. We, therefore, aimed to evaluate the safety and anti-tumour activity of fixed-duration mosunetuzumab in patients with relapsed or refractory follicular lymphoma who had received two or more previous therapies. Methods: We conducted a single-arm, multicentre, phase 2 study at 49 centres in seven countries (Australia, Canada, Germany, South Korea, Spain, UK, and USA). All patients were aged 18 years or older with histologically confirmed follicular lymphoma (grade 1–3a) and an Eastern Cooperative Oncology Group performance status of 0–1. Patients had disease that was relapsed or refractory to two or more previous lines of treatment, including an anti-CD20 therapy and an alkylating agent. Intravenous mosunetuzumab was administered in 21-day cycles with cycle 1 step-up dosing: 1 mg on cycle 1 day 1, 2 mg on cycle 1 day 8, 60 mg on cycle 1 day 15 and cycle 2 day 1, and 30 mg on day 1 of cycle 3 and onwards. Patients with a complete response by investigator assessment using the International Harmonisation Project criteria completed treatment after cycle 8, whereas patients with a partial response or stable disease continued treatment for up to 17 cycles. The primary endpoint was independent review committee-assessed complete response rate (as best response) in all enrolled patients; the primary efficacy analysis compared the observed IRC-assessed complete response rate with a 14% historical control complete response rate in a similar patient population receiving the pan class I PI3K inhibitor copanlisib. Safety was assessed in all enrolled patients. This study is registered with ClinicalTrials.gov, number NCT02500407, and is ongoing. Findings: Between May 2, 2019, and Sept 25, 2020, we enrolled 90 patients. As of the data cutoff date (Aug 27, 2021), the median follow-up was 18·3 months (IQR 13·8–23·3). According to independent review committee assessment, a complete response was recorded in 54 patients (60·0% [95% CI 49·1–70·2]). The observed complete response rate was significantly higher than the historical control complete response rate with copanlisib of 14% (p<0·0001), thereby meeting the primary study endpoint. Cytokine release syndrome was the most common adverse event (40 [44%] of 90 patients) and was predominantly grade 1 (23 [26%] of 90) and grade 2 (15 [17%]), and primarily confined to cycle 1. The most common grade 3–4 adverse events were neutropenia or neutrophil count decreased (24 [27%] of 90 patients), hypophosphataemia (15 [17%]), hyperglycaemia (seven [8%]), and anaemia (seven [8%]). Serious adverse events occurred in 42 (47%) of 90 patients. No treatment-related grade 5 (ie, fatal) adverse event occurred. Interpretation: Fixed-duration mosunetuzumab has a favourable safety profile and induces high rates of complete remissions, allowing potential administration as an outpatient regimen, in patients with relapsed or refractory follicular lymphoma and two or more previous therapies. Funding: F Hoffmann-La Roche and Genentech.