diff --git a/app/vmselect/promql/rollup.go b/app/vmselect/promql/rollup.go
index 14d9c5c08..e0ef427cf 100644
--- a/app/vmselect/promql/rollup.go
+++ b/app/vmselect/promql/rollup.go
@@ -1916,6 +1916,10 @@ func rollupChangesPrometheus(rfa *rollupFuncArg) float64 {
 	n := 0
 	for _, v := range values[1:] {
 		if v != prevValue {
+			if math.Abs(v-prevValue) < 1e-12*math.Abs(v) {
+				// This may be precision error. See https://github.com/VictoriaMetrics/VictoriaMetrics/issues/767#issuecomment-1650932203
+				continue
+			}
 			n++
 			prevValue = v
 		}
@@ -1939,6 +1943,10 @@ func rollupChanges(rfa *rollupFuncArg) float64 {
 	}
 	for _, v := range values {
 		if v != prevValue {
+			if math.Abs(v-prevValue) < 1e-12*math.Abs(v) {
+				// This may be precision error. See https://github.com/VictoriaMetrics/VictoriaMetrics/issues/767#issuecomment-1650932203
+				continue
+			}
 			n++
 			prevValue = v
 		}
@@ -1967,6 +1975,10 @@ func rollupIncreases(rfa *rollupFuncArg) float64 {
 	n := 0
 	for _, v := range values {
 		if v > prevValue {
+			if math.Abs(v-prevValue) < 1e-12*math.Abs(v) {
+				// This may be precision error. See https://github.com/VictoriaMetrics/VictoriaMetrics/issues/767#issuecomment-1650932203
+				continue
+			}
 			n++
 		}
 		prevValue = v
@@ -1998,6 +2010,10 @@ func rollupResets(rfa *rollupFuncArg) float64 {
 	n := 0
 	for _, v := range values {
 		if v < prevValue {
+			if math.Abs(v-prevValue) < 1e-12*math.Abs(v) {
+				// This may be precision error. See https://github.com/VictoriaMetrics/VictoriaMetrics/issues/767#issuecomment-1650932203
+				continue
+			}
 			n++
 		}
 		prevValue = v
diff --git a/docs/CHANGELOG.md b/docs/CHANGELOG.md
index 3b039c73a..4f7ff35cb 100644
--- a/docs/CHANGELOG.md
+++ b/docs/CHANGELOG.md
@@ -38,6 +38,7 @@ The sandbox cluster installation is running under the constant load generated by
 * FEATURE: [dashboards/all](https://grafana.com/orgs/victoriametrics): add new panel `CPU spent on GC`. It should help identifying cases when too much CPU is spent on garbage collection, and advice users on how this can be addressed.
 
 * BUGFIX: [MetricsQL](https://docs.victoriametrics.com/MetricsQL.html): properly propagate [label filters](https://docs.victoriametrics.com/keyconcepts/#filtering) from multiple arguments passed to [aggregate functions](https://docs.victoriametrics.com/metricsql/#aggregate-functions). For example, `sum({job="foo"}, {job="bar"}) by (job) + a` was improperly optimized to `sum({job="foo"}, {job="bar"}) by (job) + a{job="foo"}` before being executed. This could lead to unexpected results. See [this issue](https://github.com/VictoriaMetrics/VictoriaMetrics/issues/5604).
+* BUGFIX: [MetricsQL](https://docs.victoriametrics.com/MetricsQL.html): properly handle precision errors when calculating [changes](https://docs.victoriametrics.com/metricsql/#changes), [changes_prometheus](https://docs.victoriametrics.com/metricsql/#changes_prometheus), [increases_over_time](https://docs.victoriametrics.com/metricsql/#increases_over_time) and [resets](https://docs.victoriametrics.com/metricsql/#resets) functions. See [this issue](https://github.com/VictoriaMetrics/VictoriaMetrics/issues/767).
 * BUGFIX: all VictoriaMetrics components: consistently return 200 http status code from [`/-/reload` endpoint](https://docs.victoriametrics.com/vmagent/#configuration-update). Previously [single-node VictoriaMetrics](https://docs.victoriametrics.com/) was returning 204 http status code. See [this feature request](https://github.com/VictoriaMetrics/VictoriaMetrics/issues/5774).
 * BUGFIX: [vmui](https://docs.victoriametrics.com/#vmui): fix the graph dragging for Firefox and Safari. See [this issue](https://github.com/VictoriaMetrics/VictoriaMetrics/issues/5764).
 * BUGFIX: [vmui](https://docs.victoriametrics.com/#vmui): fix handling invalid timezone. See [this issue](https://github.com/VictoriaMetrics/VictoriaMetrics/issues/5732).