With the availability of low-cost high-quality head-mounted displays (HMDs) since 2013, there is a growing body of literature investigating the impact of immersive virtual reality (IVR) technology on education. This meta-analysis aims to synthesize the findings on the overall effects of IVR using HMDs compared to less immersive desktop virtual reality (DVR) and other traditional means of instruction. A systematic search was carried out on the literature published between 2013 and 2019. Thirty-five randomized controlled trials (RCTs) or quasi-experimental studies were identified. We conducted an analysis using the random effects model (REM) to calculate the pooled effect size. The studies were also coded to examine the moderating effects of their characteristics, such as learner stage, learning domain, learning application type, testing format, control group treatment and learning duration, on the outcome measure. The results showed that IVR using HMDs is more effective than non-immersive learning approaches with a small effect size (ES = 0.24). The key findings of the moderator analysis were that HMDs have a greater impact (a) on K-12 learners; (b) in the fields of science education and specific abilities development; (c) when offering simulation or virtual world representations; and (d) when compared with lectures or real-world practices. The meta-analysis also suggested that HMDs can improve both knowledge and skill development, and maintain the learning effect over time.