Welfare-to-work programmes have replaced passive welfare recipiency as a means of fighting poverty in many developed countries during the latest decades. There is a belief that placing welfare recipients into subsidised jobs and/or strengthening their skills and knowledge will help them acquire steady jobs. There has, however, been no systematic review of the effects of such programmes on employment, earnings and welfare payments searching systematically for studies from all parts of the world. To estimate the effects of work programmes, including elements such as job search assistance, job search training, subsidised employment, job clubs, vocational training, etc. on welfare recipients. Employment and economic self-sufficiency. We searched the following electronic databases: C2-SPECTR, Cochrane Library, MEDLINE, EMBASE, PsycINFO, Sociological Abstracts, Cinahl, Caredata, Eric, BIBSYS, SIGLE, IBSS, PAIS, and Social Science Citation Index. References from included primary reports and relevant reviews were scanned and content experts were contacted. A number of possibly relevant websites were searched. Randomised controlled trials, quasi-randomised trials, or clusterrandomised trials of welfare-to-work programmes. Studies were evaluated independently by two reviewers according to a data extraction form. The GRADE system was used for quality assessment. Outcomes on employment, earnings, welfare payments, and proportion on welfare were included in meta-analyses. Randomised controlled impact evaluations of welfare-to-work programmes came almost exclusively from the United States. A total of 46 programmes with more than 412 thousand participants were included in this review. Participants were randomised to intervention or control group, and we report follow-up outcomes from the end of the intervention and up to six years. Overall, 60.9 percent of intervention participants were employed at the follow-ups. But 57.9 percent of control participants were also employed. The random effects risk ratio (RR) for employment was 1.097 at the one-year follow-up with 95 percent confidence interval (CI) 1. 006-1.196. At two years the random effects RR was 1.092 (95% CI: 1.032-1.157), and at five years the random-effects RR was 1.037 (1.004-1.071). We estimated the overall to be 33 (95% confidence interval: 30-37). In other words, an average of 33 welfare recipients had to receive one of the work programmes in this review in order to predict that one more of them would become employed. The effect on earnings was small. At one year follow-up, the random effects Hedges. G was 0. 043 (95% CI: 0.011-0.076). At two years the random effects g was 0.044 (0.022-0.066). At five years the random-effects g was 0.011 (-0.029-0.050). The mean earnings (weighted by sample size) across all the intervention outcomes (in year 2005 US dollars) was $ 11,021 compared to $ 8,843 in the control groups. Using the Binominal Effect Size Display (BESD) this roughly corresponds to a positive impact for 51.1 percent in the intervention group and for 48.9 percent in the control group. The effect on welfare payments at one year was also small (random-effects Hedges. G =0. 038, 95% CI: -0.022-0.098). At two years, the random-effects g was 0.053 (-0.005-0.111), and at five years the fixed-effects g was 0.044 (0.028-0.060). The programmes reduced the welfare payments from $21,719 to $18,777 when averaging across all studies. Here the BESD indicated an improvement for 51.2 % of the participants in the intervention groups and for 48.8 % in the control groups. Finally, the effect on the proportion of participants on welfare after one year showed a random effects risk ratio of 0.967 (95% CI: 0.926-1.009). After two years it was of similar magnitude (random-effects RR: 0.946, 95% CI: 0.886-1.010). Finally, after five years, the fixed-effects RR was 1.003 (0.984-1.023). After taking part in a programme, 68 percent (weighted by sample size) were on welfare, compared to 72 percent in the control groups. The overall number needed to treat indicates that, on average, 27 welfare recipients (95% CI: 24-30) had to take part in a programme in order to get an additional person off welfare (overall risk ratio for all outcomes = 0.963, 95% CI: 0.948-0.978). For all four outcomes, there was significant heterogeneity which could not be sufficiently explained by moderator analysis. The GRADE quality assessment showed that for all four outcomes, the quality of evidence was very low. Welfare-to-work programmes in the USA have shown small, but consistent effects in moving welfare recipients into work, increasing earnings, and lowering welfare payments. The results are not clear for reducing the proportion of recipients receiving welfare. Little is known about the impacts of welfare-to-work programmes outside of the USA.