diff --git a/R/fcast.R b/R/fcast.R
index 3de3fcb8f..18206a106 100644
--- a/R/fcast.R
+++ b/R/fcast.R
@@ -168,10 +168,18 @@ dcast.data.table = function(data, formula, fun.aggregate = NULL, sep = "_", ...,
   setDT(dat)
 
   m = as.list(match.call()[-1L])
-  subset = m[["subset"]][[2L]]
+  subset = m[["subset"]]
   if (!is.null(subset)) {
+    if (is.call(subset) && (subset[[1L]] == quote(.) || subset[[1L]] == quote(list))) {
+      subset = subset[[2L]]
+    }
     if (is.name(subset)) subset = as.call(list(quote(`(`), subset))
-    idx = which(eval(subset, data, parent.frame())) # any advantage thro' secondary keys?
+    idx = eval(subset, data, parent.frame())
+    if (is.logical(idx) && length(idx) == 1L) {
+      idx = if (!is.na(idx) && idx) seq_len(nrow(data)) else integer(0)
+    } else {
+      idx = which(idx) # which() returns indices of TRUE, ignores FALSE and NA
+    }
     dat = .Call(CsubsetDT, dat, idx, seq_along(dat))
   }
   fun.call = m[["fun.aggregate"]]
diff --git a/inst/tests/tests.Rraw b/inst/tests/tests.Rraw
index 687bf929e..5e866cc18 100644
--- a/inst/tests/tests.Rraw
+++ b/inst/tests/tests.Rraw
@@ -21660,3 +21660,13 @@ test(2374.08, key(DT[, .(a, a)]), NULL)
 test(2374.09, key(subset(DT, select=c(a, a))), NULL)
 DT = data.table(a=1:2, a.1=3:4, val=10:11)
 test(2374.10, key(DT[, .(a.1, sum(val)), keyby=.(a, a)]), NULL)
+
+# dcast() subset supports base-style logical expressions, #4325
+DT = data.table( a = rep(1:5, 3), b = rep(1:3, each = 5), v = 1:15)
+y = data.table(a=2:5, `1`=2:5, `2`=7:10, `3`=12:15)
+setkey(y, a)
+test(2375.1, dcast(DT, a ~ b, value.var="v", subset=a > 1), y)
+test(2375.2, dcast(DT, a ~ b, value.var="v", subset=.(a > 1)), y)
+test(2375.3, identical( dcast(DT, a ~ b, value.var="v", subset=a > 2 & b != 2), dcast(DT, a ~ b, value.var="v", subset=.(a > 2 & b != 2)) ), TRUE)
+test(2375.4, dcast(DT, a ~ b, value.var="v", subset=TRUE), dcast(DT, a ~ b, value.var="v"))
+test(2375.5, nrow(dcast(DT, a ~ b, value.var="v", subset=FALSE)), 0L)