diff --git a/NEWS.md b/NEWS.md
index 1c07a5d..af0f771 100644
--- a/NEWS.md
+++ b/NEWS.md
@@ -53,6 +53,8 @@
 * `write_parquet(file = ":raw:")` now works correctly for larger data
   frames (#77).
 
+* `read_parquet()` can now read from an R connection (#71).
+
 * `read_parquet()` now reads `DECIMAL` values correctly from `INT32`
   and `INT64` columns if their `scale` is not zero.
 
diff --git a/R/read-parquet.R b/R/read-parquet.R
index 35514ef..9da621f 100644
--- a/R/read-parquet.R
+++ b/R/read-parquet.R
@@ -2,7 +2,12 @@
 #'
 #' Converts the contents of the named Parquet file to a R data frame.
 #'
-#' @param file Path to a Parquet file.
+#' @param file Path to a Parquet file. It may also be an R connection,
+#'   in which case it first reads all data from the connection, writes
+#'   it into a temporary file, then reads the temporary file, and
+#'   deletes it. The connection might be open, it which case it must be
+#'   a binary connection. If it is not open, then `read_parquet()` will
+#'   open it and also close it in the end.
 #' @param options Nanoparquet options, see [parquet_options()].
 #' @return A `data.frame` with the file's contents.
 #' @export
@@ -17,6 +22,11 @@
 #' print(str(parquet_df))
 
 read_parquet <- function(file, options = parquet_options()) {
+	if (inherits(file, "connection")) {
+		tmp <- tempfile(fileext = ".parquet")
+		dump_connection(file, tmp)
+		file <- tmp
+	}
   file <- path.expand(file)
   res <- .Call(nanoparquet_read2, file, options, sys.call())
 	dicts <- res[[2]]
@@ -46,3 +56,22 @@ read_parquet <- function(file, options = parquet_options()) {
 
   res
 }
+
+# dump the contents of a connection to path
+dump_connection <- function(con, path) {
+	if (!isOpen(con)) {
+		on.exit(close(con), add = TRUE)
+		open(con, "rb")
+	}
+	ocon <- file(path, open = "wb")
+	# 10 MB buffer by default
+	bs <- getOption("nanoparquet.con_buffer_size", 1024L * 1024L * 10)
+	while (TRUE) {
+		buf <- readBin(con, what = "raw", n = bs)
+		if (length(buf) == 0) {
+			break
+		}
+		writeBin(buf, path)
+	}
+	close(ocon)
+}
diff --git a/man/read_parquet.Rd b/man/read_parquet.Rd
index 5e74bb0..201824f 100644
--- a/man/read_parquet.Rd
+++ b/man/read_parquet.Rd
@@ -7,7 +7,12 @@
 read_parquet(file, options = parquet_options())
 }
 \arguments{
-\item{file}{Path to a Parquet file.}
+\item{file}{Path to a Parquet file. It may also be an R connection,
+in which case it first reads all data from the connection, writes
+it into a temporary file, then reads the temporary file, and
+deletes it. The connection might be open, it which case it must be
+a binary connection. If it is not open, then \code{read_parquet()} will
+open it and also close it in the end.}
 
 \item{options}{Nanoparquet options, see \code{\link[=parquet_options]{parquet_options()}}.}
 }
diff --git a/tests/testthat/test-read-parquet-connection.R b/tests/testthat/test-read-parquet-connection.R
new file mode 100644
index 0000000..b3d1974
--- /dev/null
+++ b/tests/testthat/test-read-parquet-connection.R
@@ -0,0 +1,33 @@
+test_that("not open", {
+  pf <- test_path("data/factor.parquet")
+  con <- file(pf)
+  expect_equal(
+    read_parquet(con),
+    read_parquet(pf)
+  )
+  # A closed (=invalid in R) connection will error here
+  expect_error(isOpen(con))
+})
+
+test_that("open", {
+  pf <- test_path("data/factor.parquet")
+  con <- file(pf, open = "rb")
+  on.exit(close(con), add = TRUE)
+  expect_equal(
+    read_parquet(con),
+    read_parquet(pf)
+  )
+  expect_true(isOpen(con))
+})
+
+test_that("raw, opened", {
+  pf <- test_path("data/factor.parquet")
+  bts <- readBin(pf, what = "raw", n = file.size(pf))
+  con <- rawConnection(bts, open = "rb")
+  on.exit(close(con), add = TRUE)
+  expect_equal(
+    read_parquet(con),
+    read_parquet(pf)
+  )
+  expect_true(isOpen(con))
+})