diff --git a/NEWS.md b/NEWS.md
index 431b65b4..cbcd4db9 100644
--- a/NEWS.md
+++ b/NEWS.md
@@ -1,3 +1,17 @@
+# officer 0.6.3.9000
+
+## Changes
+
+- the dataframe returned by `docx_comments()` gains a list column `para_id` containing 
+the id(s) of the parent paragraph(s). A possible breaking change is that the `commented_text` 
+is now returned as a list column to account for comments spanning multiple runs.
+The docs gain a description of the columns of the returned dataframe.
+
+## Issues
+
+- Fix. Refactor `docx_comments()` to (properly) account for comments spanning 
+multiple paragraphs, multiple comments in the same paragraph and replies. Closes #541.
+
 # officer 0.6.3
 
 ## Features
diff --git a/R/docx_comments.R b/R/docx_comments.R
index b107058f..7add121d 100644
--- a/R/docx_comments.R
+++ b/R/docx_comments.R
@@ -1,6 +1,22 @@
 #' @title Get comments in a Word document as a data.frame
 #' @description return a data.frame representing the comments in a Word document.
 #' @param x an rdocx object
+#' @details
+#' Each row of the returned data frame contains data for one comment. The
+#'    columns contain the following information:
+#' * "comment_id" - unique comment id
+#' * "author" - name of the comment author
+#' * "initials" - initials of the comment author
+#' * "date" - timestamp of the comment
+#' * "text" - a list column of characters containing the comment text. Elements can
+#'    be vectors of length > 1 if a comment contains multiple paragraphs,
+#'    blocks or runs or of length 0 if the comment is empty.
+#' * "para_id" - a list column of characters containing the parent paragraph IDs.
+#'    Elememts can be vectors of length > 1 if a comment spans multiple paragraphs
+#'    or of length 0 if the comment has no parent paragraph.
+#' * "commented_text" - a list column of characters containing the
+#'    commented text. Elememts can be vectors of length > 1 if a comment
+#'    spans multiple paragraphs or runs or of length 0 if the commented text is empty.
 #' @examples
 #' bl <- block_list(
 #'   fpar("Comment multiple words."),
@@ -10,7 +26,7 @@
 #' a_par <- fpar(
 #'   "This paragraph contains",
 #'   run_comment(
-#'   cmt = bl,
+#'     cmt = bl,
 #'     run = ftext("a comment."),
 #'     author = "Author Me",
 #'     date = "2023-06-01"
@@ -27,19 +43,33 @@
 docx_comments <- function(x) {
   stopifnot(inherits(x, "rdocx"))
 
-  comment_nodes <- xml_find_all(
-    x$doc_obj$get(), "//*[self::w:p/w:commentRangeStart]"
+  comment_ids <- xml_attr(
+    xml_find_all(
+      x$doc_obj$get(), "/w:document/w:body//*[self::w:commentRangeStart]"
+    ), "id"
   )
 
-  if (length(comment_nodes) > 0) {
-    data <- lapply(comment_nodes, comment_as_tibble)
-    data <- rbind_match_columns(data)
-  } else {
-    data <- data.frame(
-      comment_id = integer(0),
-      commented_text = character(0)
+  comment_text_runs <- lapply(comment_ids, \(id) {
+    xml_find_all(
+      x$doc_obj$get(),
+      paste0(
+        "/w:document/w:body//*[self::w:r[w:t and",
+        "preceding::w:commentRangeStart[@w:id=\'", id, "\']",
+        " and ",
+        "following::w:commentRangeEnd[@w:id=\'", id, "\']]]"
+      )
     )
-  }
+  })
+
+  data <- data.frame(
+    comment_id = comment_ids
+  )
+  # Add parent paragraph id
+  data$para_id <- lapply(
+    comment_text_runs,
+    function(x) xml_attr(xml_parent(x), "paraId")
+  )
+  data$commented_text <- lapply(comment_text_runs, xml_text)
 
   comments <- xml_find_all(x$comments$get(), "//w:comments/w:comment")
 
@@ -57,30 +87,5 @@ docx_comments <- function(x) {
   )
 
   data <- merge(out, data, by = "comment_id", all.x = TRUE)
-
-  data
-}
-
-comment_as_tibble <- function(node) {
-  node_name <- xml_name(node)
-  name_children <- xml_name(xml_children(node))
-
-  comment_range <- grep("commentRange", name_children)
-
-  comment_data <- data.frame(
-    comment_id = xml_attr(xml_child(node, comment_range[[1]]), "id"),
-    stringsAsFactors = FALSE
-  )
-  comment_range <- seq(comment_range[[1]] + 1, comment_range[[2]] - 1)
-  comment_data$commented_text <-
-    paste0(
-      vapply(
-        comment_range,
-        function(x) xml_text(xml_child(node, x)),
-        character(1)
-      ),
-      collapse = ""
-    )
-
-  comment_data
+  data[order(as.integer(data$comment_id)), ]
 }
diff --git a/man/docx_comments.Rd b/man/docx_comments.Rd
index e49bd1a6..60662e62 100644
--- a/man/docx_comments.Rd
+++ b/man/docx_comments.Rd
@@ -12,6 +12,25 @@ docx_comments(x)
 \description{
 return a data.frame representing the comments in a Word document.
 }
+\details{
+Each row of the returned data frame contains data for one comment. The
+columns contain the following information:
+\itemize{
+\item "comment_id" - unique comment id
+\item "author" - name of the comment author
+\item "initials" - initials of the comment author
+\item "date" - timestamp of the comment
+\item "text" - a list column of characters containing the comment text. Elements can
+be vectors of length > 1 if a comment contains multiple paragraphs,
+blocks or runs or of length 0 if the comment is empty.
+\item "para_id" - a list column of characters containing the parent paragraph IDs.
+Elememts can be vectors of length > 1 if a comment spans multiple paragraphs
+or of length 0 if the comment has no parent paragraph.
+\item "commented_text" - a list column of characters containing the
+commented text. Elememts can be vectors of length > 1 if a comment
+spans multiple paragraphs or runs or of length 0 if the commented text is empty.
+}
+}
 \examples{
 bl <- block_list(
   fpar("Comment multiple words."),
@@ -21,7 +40,7 @@ bl <- block_list(
 a_par <- fpar(
   "This paragraph contains",
   run_comment(
-  cmt = bl,
+    cmt = bl,
     run = ftext("a comment."),
     author = "Author Me",
     date = "2023-06-01"
diff --git a/tests/testthat/docs_dir/test-docx_comments.docx b/tests/testthat/docs_dir/test-docx_comments.docx
new file mode 100644
index 00000000..f4bbb3f2
Binary files /dev/null and b/tests/testthat/docs_dir/test-docx_comments.docx differ
diff --git a/tests/testthat/test-docx-comments.R b/tests/testthat/test-docx-comments.R
index 6ff56f64..82ce04ac 100644
--- a/tests/testthat/test-docx-comments.R
+++ b/tests/testthat/test-docx-comments.R
@@ -50,3 +50,262 @@ test_that("add comments", {
   expect_length(xml_children(comment1), 2)
   expect_length(xml_children(comment2), 1)
 })
+
+test_that("docx_comments accounts for multiple comments in a paragraph", {
+  multi_comment_par <- fpar(
+    "This paragraph",
+    run_comment(
+      cmt = block_list(
+        fpar(
+          ftext("First Comment.")
+        )
+      ),
+      run = ftext("contains"),
+      author = "Author Me",
+      date = "2023-06-01"
+    ),
+    "multiple",
+    run_comment(
+      cmt = block_list(
+        fpar(
+          ftext("Second Comment.")
+        )
+      ),
+      run = ftext("comments"),
+      author = "Author Me",
+      date = "2023-06-01"
+    )
+  )
+
+  doc <- read_docx()
+  doc <- body_add_fpar(doc, value = multi_comment_par, style = "Normal")
+  docx_file <- print(doc, target = tempfile(fileext = ".docx"))
+
+  comments <- docx_comments(read_docx(docx_file))
+
+  expect_equal(nrow(comments), 2)
+
+  expect_equal(
+    comments$text,
+    list("First Comment.", "Second Comment.")
+  )
+  expect_equal(
+    comments$commented_text,
+    list("contains", "comments")
+  )
+})
+
+test_that("docx_comments", {
+  example_docx <- "docs_dir/test-docx_comments.docx"
+  doc <- read_docx(path = example_docx)
+
+  comments <- docx_comments(doc)
+
+  # Doc includes 16 comments
+  expect_equal(nrow(comments), 16)
+  # No NAs in "commented_text"
+  expect_true(
+    all(
+      vapply(
+        comments[["commented_text"]],
+        function(x) all(!is.na(x)),
+        FUN.VALUE = logical(1)
+      )
+    )
+  )
+  # Accounts for empty comments or multi line comments
+  expect_true(
+    all(
+      lengths(comments[["text"]][-c(1, 4)]) == 1
+    )
+  )
+  ## Comment 1 has 2 lines
+  expect_equal(
+    length(comments[["text"]][[1]]), 2
+  )
+  ## Comment 4 is empty
+  expect_identical(
+    comments[["text"]][[4]], character(0)
+  )
+})
+
+test_that("docx_comments accounts for comments spanning no or multiple paragraphs", {
+  example_docx <- "docs_dir/test-docx_comments.docx"
+  doc <- read_docx(path = example_docx)
+
+  comments <- docx_comments(doc)
+
+  expect_true(
+    all(
+      lengths(comments[["para_id"]][-c(5, 6)]) == 1
+    )
+  )
+  ## Comment 5 spans no paragraph
+  expect_identical(
+    comments[["para_id"]][[5]], character(0)
+  )
+  expect_equal(
+    paste(comments[["commented_text"]][[5]], collapse = " "),
+    ""
+  )
+  ## Comment 6 spans 2 paragraphs
+  expect_equal(
+    length(comments[["para_id"]][[6]]), 2
+  )
+  expect_equal(
+    paste(comments[["commented_text"]][[6]], collapse = " "),
+    "a comment … … which spans multiple paragraphs."
+  )
+})
+
+test_that("docx_comments accounts for comments spanning no or multiple runs", {
+  example_docx <- "docs_dir/test-docx_comments.docx"
+  doc <- read_docx(path = example_docx)
+
+  comments <- docx_comments(doc)
+
+  expect_true(
+    all(
+      lengths(comments[["commented_text"]][-c(5, 6, 7, 15)]) == 1
+    )
+  )
+  ## Comment 5 spans no run
+  expect_identical(
+    comments[["commented_text"]][[5]], character(0)
+  )
+  expect_equal(
+    paste(comments[["commented_text"]][[5]], collapse = " "),
+    ""
+  )
+  ## Comment 6 spans 2 runs as it spans 2 paragraphs
+  expect_equal(
+    length(comments[["commented_text"]][[6]]), 2
+  )
+  expect_equal(
+    paste(comments[["commented_text"]][[6]], collapse = " "),
+    "a comment … … which spans multiple paragraphs."
+  )
+  ## Comment 7 spans 3 runs.
+  expect_equal(
+    length(comments[["commented_text"]][[7]]), 3
+  )
+  expect_equal(
+    paste(comments[["commented_text"]][[7]], collapse = ""),
+    "spanning multiple runs."
+  )
+  ## Comment 15 spans 3 runs because of an inner comment
+  expect_equal(
+    length(comments[["commented_text"]][[15]]), 3
+  )
+  expect_equal(
+    paste(comments[["commented_text"]][[15]], collapse = ""),
+    "This paragraph contains two nested comments."
+  )
+})
+
+test_that("docx_comments accounts for replies", {
+  example_docx <- "docs_dir/test-docx_comments.docx"
+  doc <- read_docx(path = example_docx)
+
+  comments <- docx_comments(doc)
+
+  # Make "unique" id based on commented text and paragraph id
+  comments$unique_id <- paste(
+    comments$para_id, comments$commented_text,
+    sep = "."
+  )
+  comments$unique_id <- factor(
+    comments$unique_id,
+    levels = unique(comments$unique_id)
+  )
+  comments$unique_id <- as.integer(comments$unique_id)
+
+  comments_split <- split(
+    comments,
+    comments$unique_id
+  )
+
+  # Accouting for replies we have only 12 comments
+  expect_equal(
+    length(comments_split),
+    12
+  )
+  expect_equal(
+    vapply(
+      comments_split[-c(8, 9, 10)], nrow,
+      FUN.VALUE = integer(1), USE.NAMES = FALSE
+    ),
+    rep(1, 9)
+  )
+  ## 8 has two replies
+  expect_equal(
+    nrow(comments_split[[8]]),
+    3
+  )
+  expect_equal(
+    unique(unlist(comments[8:10, "commented_text"])),
+    comments[["commented_text"]][[8]]
+  )
+  expect_equal(
+    unique(unlist(comments[8:10, "para_id"])),
+    comments[["para_id"]][[8]]
+  )
+  ## 9 and 10 have one reply each
+  expect_equal(
+    nrow(comments_split[[9]]),
+    2
+  )
+  expect_equal(
+    unique(unlist(comments[11:12, "commented_text"])),
+    comments[["commented_text"]][[11]]
+  )
+  expect_equal(
+    unique(unlist(comments[11:12, "para_id"])),
+    comments[["para_id"]][[11]]
+  )
+  expect_equal(
+    nrow(comments_split[[10]]),
+    2
+  )
+  expect_equal(
+    unique(unlist(comments[13:14, "commented_text"])),
+    comments[["commented_text"]][[13]]
+  )
+  expect_equal(
+    unique(unlist(comments[13:14, "para_id"])),
+    comments[["para_id"]][[13]]
+  )
+})
+
+test_that("docx_comments accounts for nested comments", {
+  example_docx <- "docs_dir/test-docx_comments.docx"
+  doc <- read_docx(path = example_docx)
+
+  comments <- docx_comments(doc)
+
+  ## Outer Comment 15 spans 3 runs
+  expect_equal(
+    length(comments[["commented_text"]][[15]]), 3
+  )
+  expect_equal(
+    paste(comments[["commented_text"]][[15]], collapse = ""),
+    "This paragraph contains two nested comments."
+  )
+  expect_equal(
+    paste(comments[["text"]][[15]], collapse = ""),
+    "Outer Comment."
+  )
+
+  ## Inner Comment 16 spans 1 run
+  expect_equal(
+    length(comments[["commented_text"]][[16]]), 1
+  )
+  expect_equal(
+    paste(comments[["commented_text"]][[16]], collapse = ""),
+    "contains two "
+  )
+  expect_equal(
+    paste(comments[["text"]][[16]], collapse = ""),
+    "Inner Comment."
+  )
+})