Influenza H5N1 is thought to be a likely causative agent for a future human influenza pandemic. Several types of H5N1 vaccine have been tested, including different doses and adjuvants, and a meta-analysis is needed to identify the best formulation. We searched Medline, Embase, the Cochrane Library, and other online databases to February, 2009, in any language for randomised trials comparing different H5N1 vaccines with or without placebo in healthy adults. Primary outcomes were seroconversion, seroresponse, or both according to haemagglutination-inhibition and microneutralisation. Secondary outcomes were adverse events. Because of the large number of compared formulations, multiple-treatments meta-analysis was used for primary outcomes. Direct-comparison meta-analyses were also done. We included 13 trials, which assessed 58 groups. With non-aluminium adjuvant, sufficiently high immunogenicity (greater than 70%) was achieved even at 12 microg or less (given as two doses of 6 microg or less), and higher doses did not provide major improvements. Immunogenicity for non-adjuvanted and aluminium-adjuvanted formulations increased with increasing dose, but was not sufficiently high. No serious vaccine-related adverse events were reported across 9600 participants. Currently, H5N1 influenza vaccines with non-aluminium adjuvants might represent the best available option in a pandemic. Large-scale studies are needed to verify the high immunogenicity of non-aluminium-adjuvanted vaccines that use very low doses of antigen.