BACKGROUND: Few data exist on the comparative safety and immunogenicity of different COVID-19 vaccines given as a third (booster) dose. To generate data to optimise selection of booster vaccines, we investigated the reactogenicity and immunogenicity of seven different COVID-19 vaccines as a third dose after two doses of ChAdOx1 nCov-19 (Oxford-AstraZeneca; hereafter referred to as ChAd) or BNT162b2 (Pfizer-BioNtech, hearafter referred to as BNT). METHODS: COV-BOOST is a multicentre, randomised, controlled, phase 2 trial of third dose booster vaccination against COVID-19. Participants were aged older than 30 years, and were at least 70 days post two doses of ChAd or at least 84 days post two doses of BNT primary COVID-19 immunisation course, with no history of laboratory-confirmed SARS-CoV-2 infection. 18 sites were split into three groups (A, B, and C). Within each site group (A, B, or C), participants were randomly assigned to an experimental vaccine or control. Group A received NVX-CoV2373 (Novavax; hereafter referred to as NVX), a half dose of NVX, ChAd, or quadrivalent meningococcal conjugate vaccine (MenACWY) control (1:1:1:1). Group B received BNT, VLA2001 (Valneva; hereafter referred to as VLA), a half dose of VLA, Ad26.COV2.S (Janssen; hereafter referred to as Ad26) or MenACWY (1:1:1:1:1). Group C received mRNA1273 (Moderna; hereafter referred to as m1273), CVnCov (CureVac; hereafter referred to as CVn), a half dose of BNT, or MenACWY (1:1:1:1). Participants and all investigatory staff were blinded to treatment allocation. Coprimary outcomes were safety and reactogenicity and immunogenicity of anti-spike IgG measured by ELISA. The primary analysis for immunogenicity was on a modified intention-to-treat basis; safety and reactogenicity were assessed in the intention-to-treat population. Secondary outcomes included assessment of viral neutralisation and cellular responses. This trial is registered with ISRCTN, number 73765130. FINDINGS: Between June 1 and June 30, 2021, 3498 people were screened. 2878 participants met eligibility criteria and received COVID-19 vaccine or control. The median ages of ChAd/ChAd-primed participants were 53 years (IQR 44-61) in the younger age group and 76 years (73-78) in the older age group. In the BNT/BNT-primed participants, the median ages were 51 years (41-59) in the younger age group and 78 years (75-82) in the older age group. In the ChAd/ChAD-primed group, 676 (46·7%) participants were female and 1380 (95·4%) were White, and in the BNT/BNT-primed group 770 (53·6%) participants were female and 1321 (91·9%) were White. Three vaccines showed overall increased reactogenicity: m1273 after ChAd/ChAd or BNT/BNT; and ChAd and Ad26 after BNT/BNT. For ChAd/ChAd-primed individuals, spike IgG geometric mean ratios (GMRs) between study vaccines and controls ranged from 1·8 (99% CI 1·5-2·3) in the half VLA group to 32·3 (24·8-42·0) in the m1273 group. GMRs for wild-type cellular responses compared with controls ranged from 1·1 (95% CI 0·7-1·6) for ChAd to 3·6 (2·4-5·5) for m1273. For BNT/BNT-primed individuals, spike IgG GMRs ranged from 1·3 (99% CI 1·0-1·5) in the half VLA group to 11·5 (9·4-14·1) in the m1273 group. GMRs for wild-type cellular responses compared with controls ranged from 1·0 (95% CI 0·7-1·6) for half VLA to 4·7 (3·1-7·1) for m1273. The results were similar between those aged 30-69 years and those aged 70 years and older. Fatigue and pain were the most common solicited local and systemic adverse events, experienced more in people aged 30-69 years than those aged 70 years or older. Serious adverse events were uncommon, similar in active vaccine and control groups. In total, there were 24 serious adverse events: five in the control group (two in control group A, three in control group B, and zero in control group C), two in Ad26, five in VLA, one in VLA-half, one in BNT, two in BNT-half, two in ChAd, one in CVn, two in NVX, two in NVX-half, and one in m1273. INTERPRETATION: All study vaccines boosted antibody and neutralising responses after ChAd/ChAd initial course and all except one after BNT/BNT, with no safety concerns. Substantial differences in humoral and cellular responses, and vaccine availability will influence policy choices for booster vaccination. FUNDING: UK Vaccine Taskforce and National Institute for Health Research.