diff --git a/r/NAMESPACE b/r/NAMESPACE
index ab45aa9985e..814868d8ade 100644
--- a/r/NAMESPACE
+++ b/r/NAMESPACE
@@ -327,6 +327,7 @@ importFrom(rlang,quos)
 importFrom(rlang,seq2)
 importFrom(rlang,set_names)
 importFrom(rlang,syms)
+importFrom(rlang,trace_back)
 importFrom(rlang,warn)
 importFrom(stats,median)
 importFrom(stats,na.exclude)
diff --git a/r/R/metadata.R b/r/R/metadata.R
index 408c2214a31..505d0653b4a 100644
--- a/r/R/metadata.R
+++ b/r/R/metadata.R
@@ -50,6 +50,7 @@
   })
 }
 
+#' @importFrom rlang trace_back
 apply_arrow_r_metadata <- function(x, r_metadata) {
   tryCatch({
     columns_metadata <- r_metadata$columns
@@ -60,9 +61,27 @@ apply_arrow_r_metadata <- function(x, r_metadata) {
         }
       }
     } else if (is.list(x) && !inherits(x, "POSIXlt") && !is.null(columns_metadata)) {
-      x <- map2(x, columns_metadata, function(.x, .y) {
-        apply_arrow_r_metadata(.x, .y)
-      })
+      # If we have a list and "columns_metadata" this applies row-level metadata
+      # inside of a column in a dataframe.
+
+      # However, if we are inside of a dplyr collection (including all datasets),
+      # we cannot apply this row-level metadata, since the order of the rows is
+      # not guaranteed to be the same, so don't even try, but warn what's going on
+      trace <- trace_back()
+      in_dplyr_collect <- any(map_lgl(trace$calls, function(x) {
+        grepl("collect.arrow_dplyr_query", x, fixed = TRUE)[[1]]
+      }))
+      if (in_dplyr_collect) {
+        warning(
+          "Row-level metadata is not compatible with this operation and has ",
+          "been ignored",
+          call. = FALSE
+        )
+      } else {
+        x <- map2(x, columns_metadata, function(.x, .y) {
+          apply_arrow_r_metadata(.x, .y)
+        })
+      }
       x
     }
 
@@ -116,9 +135,23 @@ arrow_attributes <- function(x, only_top_level = FALSE) {
 
   columns <- NULL
   if (is.list(x) && !inherits(x, "POSIXlt")) {
-    # for list columns, we also keep attributes of each
-    # element in columns
-    columns <- map(x, arrow_attributes)
+    # However, if we are inside of a dplyr collection (including all datasets),
+    # we cannot apply this row-level metadata, since the order of the rows is
+    # not guaranteed to be the same, so don't even try, but warn what's going on
+    trace <- trace_back()
+    in_dataset_write <- any(map_lgl(trace$calls, function(x) {
+      grepl("write_dataset", x, fixed = TRUE)[[1]]
+    }))
+    if (in_dataset_write) {
+      warning(
+        "Row-level metadata is not compatible with datasets and will be discarded",
+        call. = FALSE
+      )
+    } else {
+      # for list columns, we also keep attributes of each
+      # element in columns
+      columns <- map(x, arrow_attributes)
+    }
     if (all(map_lgl(columns, is.null))) {
       columns <- NULL
     }
diff --git a/r/tests/testthat/helper-arrow.R b/r/tests/testthat/helper-arrow.R
index 0abbfb6a13a..5f2dad841a1 100644
--- a/r/tests/testthat/helper-arrow.R
+++ b/r/tests/testthat/helper-arrow.R
@@ -67,3 +67,9 @@ test_that <- function(what, code) {
 r_only <- function(code) {
   withr::with_options(list(..skip.tests = FALSE), code)
 }
+
+make_temp_dir <- function() {
+  path <- tempfile()
+  dir.create(path)
+  normalizePath(path, winslash = "/")
+}
diff --git a/r/tests/testthat/test-dataset.R b/r/tests/testthat/test-dataset.R
index a0b1bdae022..66493376e74 100644
--- a/r/tests/testthat/test-dataset.R
+++ b/r/tests/testthat/test-dataset.R
@@ -21,12 +21,6 @@ context("Dataset")
 
 library(dplyr)
 
-make_temp_dir <- function() {
-  path <- tempfile()
-  dir.create(path)
-  normalizePath(path, winslash = "/")
-}
-
 dataset_dir <- make_temp_dir()
 hive_dir <- make_temp_dir()
 ipc_dir <- make_temp_dir()
diff --git a/r/tests/testthat/test-metadata.R b/r/tests/testthat/test-metadata.R
index afce1c2244c..de3542b1c60 100644
--- a/r/tests/testthat/test-metadata.R
+++ b/r/tests/testthat/test-metadata.R
@@ -205,3 +205,45 @@ test_that("metadata of list elements (ARROW-10386)", {
   expect_identical(attr(as.data.frame(tab)$x[[1]], "foo"), "bar")
   expect_identical(attr(as.data.frame(tab)$x[[2]], "baz"), "qux")
 })
+
+
+test_that("metadata of list elements (ARROW-10386)", {
+  skip_if_not_available("dataset")
+  skip_if_not_available("parquet")
+
+  library(dplyr)
+
+  df <- tibble::tibble(
+    metadata = list(
+      structure(1, my_value_as_attr = 1),
+      structure(2, my_value_as_attr = 2),
+      structure(3, my_value_as_attr = 3),
+      structure(4, my_value_as_attr = 3)),
+    int = 1L:4L,
+    part = c(1, 3, 2, 1)
+  )
+
+  dst_dir <- make_temp_dir()
+  expect_warning(
+    write_dataset(df, dst_dir, partitioning = "part"),
+    "Row-level metadata is not compatible with datasets and will be discarded"
+  )
+
+  # but we need to write a dataset with row-level metadata to make sure when
+  # reading ones that have been written with them we warn appropriately
+  fake_func_name <- write_dataset
+  fake_func_name(df, dst_dir, partitioning = "part")
+
+  ds <- open_dataset(dst_dir)
+  expect_warning(
+    df_from_ds <- collect(ds),
+    "Row-level metadata is not compatible with this operation and has been ignored"
+  )
+  expect_equal(df_from_ds[c(1, 4, 3, 2), ], df, check.attributes = FALSE)
+
+  # however there is *no* warning if we don't select the metadata column
+  expect_warning(
+    df_from_ds <- ds %>% select(int) %>% collect(),
+    NA
+  )
+})