diff --git a/NEWS.md b/NEWS.md
index 6aae9977..97a64dbe 100644
--- a/NEWS.md
+++ b/NEWS.md
@@ -1,5 +1,6 @@
 # stringr (development version)
 
+* `str_sub<-` now gives a more informative error if `value` is not the correct length.
 * `str_*` now errors if `pattern` includes any `NA`s (@nash-delcamp-slp, #546).
 * `str_view()` now displays a message when called with a zero-length character
   vector (@LouisMPenrod, #497).
diff --git a/R/sub.R b/R/sub.R
index 25a9c80b..961ac94e 100644
--- a/R/sub.R
+++ b/R/sub.R
@@ -6,18 +6,25 @@
 #'
 #' @inheritParams str_detect
 #' @param start,end A pair of integer vectors defining the range of characters
-#'   to extract (inclusive).
+#'   to extract (inclusive). Positive values count from the left of the string,
+#'   and negative values count from the right. In other words, if `string` is
+#'   `"abcdef"` then 1 refers to `"a"` and -1 refers to `"f"`.
 #'
 #'   Alternatively, instead of a pair of vectors, you can pass a matrix to
 #'   `start`. The matrix should have two columns, either labelled `start`
-#'   and `end`, or `start` and `length`.
+#'   and `end`, or `start` and `length`. This makes `str_sub()` work directly
+#'   with the output from [str_locate()] and friends.
+#'
 #' @param omit_na Single logical value. If `TRUE`, missing values in any of the
 #'   arguments provided will result in an unchanged input.
-#' @param value replacement string
+#' @param value Replacement string.
 #' @return
 #' * `str_sub()`: A character vector the same length as `string`/`start`/`end`.
 #' * `str_sub_all()`: A list the same length as `string`. Each element is
 #'    a character vector the same length as `start`/`end`.
+#'
+#' If `end` comes before `start` or `start` is outside the range of `string`
+#' then the corresponding output will be the empty string.
 #' @seealso The underlying implementation in [stringi::stri_sub()]
 #' @export
 #' @examples
@@ -28,7 +35,7 @@
 #' str_sub(hw, 8, 14)
 #' str_sub(hw, 8)
 #'
-#' # Negative indices index from end of string
+#' # Negative values index from end of string
 #' str_sub(hw, -1)
 #' str_sub(hw, -7)
 #' str_sub(hw, end = -7)
@@ -67,8 +74,8 @@ str_sub <- function(string, start = 1L, end = -1L) {
 
 #' @export
 #' @rdname str_sub
-"str_sub<-" <- function(string, start = 1L, end = -1L, omit_na = FALSE,  value) {
-  vctrs::vec_size_common(string = string, start = start, end = end)
+"str_sub<-" <- function(string, start = 1L, end = -1L, omit_na = FALSE, value) {
+  vctrs::vec_size_common(string = string, start = start, end = end, value = value)
 
   if (is.matrix(start)) {
     stri_sub(string, from = start, omit_na = omit_na) <- value
diff --git a/man/str_sub.Rd b/man/str_sub.Rd
index a329e7ad..9e5b62ba 100644
--- a/man/str_sub.Rd
+++ b/man/str_sub.Rd
@@ -17,16 +17,19 @@ str_sub_all(string, start = 1L, end = -1L)
 coercible to one.}
 
 \item{start, end}{A pair of integer vectors defining the range of characters
-to extract (inclusive).
+to extract (inclusive). Positive values count from the left of the string,
+and negative values count from the right. In other words, if \code{string} is
+\code{"abcdef"} then 1 refers to \code{"a"} and -1 refers to \code{"f"}.
 
 Alternatively, instead of a pair of vectors, you can pass a matrix to
 \code{start}. The matrix should have two columns, either labelled \code{start}
-and \code{end}, or \code{start} and \code{length}.}
+and \code{end}, or \code{start} and \code{length}. This makes \code{str_sub()} work directly
+with the output from \code{\link[=str_locate]{str_locate()}} and friends.}
 
 \item{omit_na}{Single logical value. If \code{TRUE}, missing values in any of the
 arguments provided will result in an unchanged input.}
 
-\item{value}{replacement string}
+\item{value}{Replacement string.}
 }
 \value{
 \itemize{
@@ -34,6 +37,9 @@ arguments provided will result in an unchanged input.}
 \item \code{str_sub_all()}: A list the same length as \code{string}. Each element is
 a character vector the same length as \code{start}/\code{end}.
 }
+
+If \code{end} comes before \code{start} or \code{start} is outside the range of \code{string}
+then the corresponding output will be the empty string.
 }
 \description{
 \code{str_sub()} extracts or replaces the elements at a single position in each
@@ -48,7 +54,7 @@ str_sub(hw, end = 6)
 str_sub(hw, 8, 14)
 str_sub(hw, 8)
 
-# Negative indices index from end of string
+# Negative values index from end of string
 str_sub(hw, -1)
 str_sub(hw, -7)
 str_sub(hw, end = -7)
diff --git a/tests/testthat/_snaps/sub.md b/tests/testthat/_snaps/sub.md
new file mode 100644
index 00000000..4b76a065
--- /dev/null
+++ b/tests/testthat/_snaps/sub.md
@@ -0,0 +1,13 @@
+# bad vectorisation gives informative error
+
+    Code
+      str_sub(x, 1:2, 1:3)
+    Condition
+      Error in `str_sub()`:
+      ! Can't recycle `string` (size 2) to match `end` (size 3).
+    Code
+      str_sub(x, 1:2, 1:2) <- 1:3
+    Condition
+      Error in `str_sub<-`:
+      ! Can't recycle `string` (size 2) to match `value` (size 3).
+
diff --git a/tests/testthat/test-sub.R b/tests/testthat/test-sub.R
index e33de87e..5ad31d8f 100644
--- a/tests/testthat/test-sub.R
+++ b/tests/testthat/test-sub.R
@@ -73,7 +73,11 @@ test_that("missing arguments give missing results", {
 
   expect_equal(str_sub("test", NA, NA), NA_character_)
   expect_equal(str_sub(c(NA, "test"), NA, NA), rep(NA_character_, 2))
+})
 
+test_that("negative length or out of range gives empty string", {
+  expect_equal(str_sub("abc", 2, 1), "")
+  expect_equal(str_sub("abc", 4, 5), "")
 })
 
 test_that("replacement works", {
@@ -101,3 +105,11 @@ test_that("replacement with NA works", {
   str_sub(x, 1, 1, omit_na = TRUE) <- NA
   expect_equal(x, "BBCDEF")
 })
+
+test_that("bad vectorisation gives informative error", {
+  x <- "a"
+  expect_snapshot(error = TRUE, {
+    str_sub(x, 1:2, 1:3)
+    str_sub(x, 1:2, 1:2) <- 1:3
+  })
+})