A forest plot is the graphical summary of a meta-analysis: it stacks the result of every included study in a single column, draws each study's effect estimate as a square with a horizontal line for its confidence interval, and places a diamond at the bottom for the pooled effect that combines them all. Reading one well means you can judge, at a glance, the direction and size of an effect, how precise each study is, how much the studies agree, and whether the combined evidence points to a real difference. This guide walks through every element on the figure in the order you should actually scan it, so you can move from raw marks to a defensible conclusion.
Start with the line of no effect
Before you read a single study, find the vertical reference line running down the middle of the plot. This is the line of no effect, and its position depends on the effect measure. For ratio measures such as an odds ratio or a risk ratio, the line sits at 1, because a ratio of 1 means the event is equally likely in both groups. For difference measures such as a mean difference or a standardized mean difference, the line sits at 0, because a difference of zero means the two groups performed identically. Everything else on the plot is interpreted relative to this line, so anchoring to it first keeps you from misreading the whole figure. If you are unsure which ratio applies to your data, our note on when to use an odds ratio or a risk ratio explains when each is appropriate.
Note which side of the line favours which group. Most plots label this directly beneath the axis, for example "favours treatment" on the left and "favours control" on the right. Reading these labels once prevents the common error of reporting an effect in the wrong direction.
Read each study row: the square and the line
Each horizontal row represents one study, usually labelled with the first author and year. The square on that row marks the study's point estimate, the single best estimate of the effect from that study alone. Where the square falls relative to the line of no effect tells you the direction of that study's result.
Square size encodes study weight
The size of each square is not decorative. A larger square means the study carries more weight in the pooled result, and weight is driven mainly by precision, which in turn reflects sample size and the number of events. A large multicentre trial will show a big square and pull the diamond toward its own estimate; a small pilot study will show a tiny square and barely move the pooled value. When you scan the column, the big squares are the studies actually deciding the answer.
The horizontal line is the confidence interval
The line extending left and right from each square is the study's confidence interval, almost always the 95 percent interval. A short line means a precise estimate; a long line means the study is small or noisy and its true effect could lie across a wide range. The single most important check on each row is whether that interval crosses the line of no effect. If the confidence interval crosses the reference line, that individual study is not statistically significant on its own, because its range of plausible values includes "no effect." If the whole interval sits to one side of the line, that study shows a significant effect in that direction.
- Interval entirely left of the line: a significant effect favouring whichever group the left side represents.
- Interval entirely right of the line: a significant effect in the opposite direction.
- Interval touching or crossing the line: that study cannot rule out no effect on its own.
What overlap between intervals tells you
Once you have read the rows individually, step back and look at how the horizontal lines line up with each other. When most confidence intervals overlap heavily and sit on the same side of the reference line, the studies are telling a consistent story, and you can be more comfortable combining them. When the intervals barely overlap, or when some sit firmly left of the line while others sit firmly right, the studies disagree, and that visual scatter is your first warning of heterogeneity. The plot lets you see this before you ever read a statistic: a tidy vertical stack of overlapping lines looks very different from a fan of intervals pointing in opposite directions.
The pooled summary diamond
At the foot of the plot sits the summary diamond, the headline result of the meta-analysis. Its horizontal centre marks the pooled effect estimate, the weighted average of every study, and the left and right tips of the diamond mark the confidence interval around that pooled value. Read the diamond exactly as you read a study row, but with more authority, because it represents the combined evidence.
- Position: where the centre of the diamond falls tells you the average direction and magnitude of the effect across all studies.
- Width: a narrow diamond means a precise pooled estimate, usually because the included studies were large or numerous; a wide diamond signals remaining uncertainty.
- Relationship to the line of no effect: if the diamond clears the reference line entirely, the pooled effect is statistically significant; if any part of the diamond touches or crosses the line, the combined evidence cannot exclude no effect.
The diamond's position is also shaped by which model produced it. A fixed-effect and a random-effects model can place the diamond in noticeably different spots and give it a different width, so always check which model the analysis used before you quote the number. You can experiment with both models directly in our model-comparison calculator to see how the diamond shifts.
The heterogeneity statistics beneath the plot
The numbers printed under the diamond quantify the disagreement your eye already spotted. You will typically see Cochran's Q with its p-value, the I-squared statistic, and often tau-squared. Cochran's Q tests whether the observed scatter is more than chance; a small p-value flags real between-study variation. I-squared then expresses, as a percentage, how much of the total variation comes from genuine differences between studies rather than from sampling error.
As a rough guide, an I-squared near 25 percent is low, around 50 percent is moderate, and 75 percent or higher is considerable heterogeneity that should make you cautious about a single pooled number. High I-squared does not invalidate the analysis, but it tells you the average effect hides meaningful variation, and it pushes you toward a random-effects model, subgroup analysis, or a prediction interval. Our deeper explainer on what the I-squared statistic is telling you covers how to act on each band.
Checking for small-study effects
A forest plot shows you the effects and their consistency, but it does not reveal whether small studies are systematically missing or biased. Before you trust a pooled result built from many small trials, pair the forest plot with a publication bias funnel plot. Asymmetry there can explain why a forest plot's smaller studies cluster on one side, and it changes how confidently you should report the diamond.
Drawing a conclusion from the whole plot
Put the elements together in sequence. First, read the diamond: its position gives the direction and size of the average effect, and whether it clears the line of no effect tells you if that effect is significant. Second, read the I-squared and the spread of the study intervals to judge how much faith to place in that single average. A diamond well clear of the line with low heterogeneity is a strong, consistent result. A diamond clear of the line but with high heterogeneity is a real effect whose size varies across settings, so report it with the variation in mind. A diamond straddling the line means the evidence, pooled, does not demonstrate an effect, regardless of how a few individual studies looked.
That disciplined read is the difference between describing a figure and interpreting evidence, and it is exactly what peer reviewers expect to see in the results narrative of a well-conducted meta-analysis. When you are ready to build your own figure from extracted data, you can draw a forest plot from your extracted data and apply every check in this guide to your own studies.