diff --git a/R/loo.R b/R/loo.R
index 2565bee56..3789da20c 100644
--- a/R/loo.R
+++ b/R/loo.R
@@ -27,9 +27,10 @@
 #'  details.
 #' @param reloo Logical; Indicate whether \code{\link{reloo}}
 #'  should be applied on problematic observations. Defaults to \code{FALSE}.
-#' @param k_threshold The threshold at which pareto \eqn{k}
-#'   estimates are treated as problematic. Defaults to \code{0.7}.
-#'   Only used if argument \code{reloo} is \code{TRUE}.
+#' @param k_threshold The Pareto \eqn{k} threshold for which observations
+#'   \code{\link{loo_moment_match}} or \code{\link{reloo}} is applied if
+#'   argument \code{moment_match} or \code{reloo} is \code{TRUE}.
+#'   Defaults to \code{0.7}.
 #'   See \code{\link[loo:pareto-k-diagnostic]{pareto_k_ids}} for more details.
 #' @param save_psis Should the \code{"psis"} object created internally be saved
 #'   in the returned object? For more details see \code{\link[loo:loo]{loo}}.
@@ -677,13 +678,18 @@ recommend_loo_options <- function(loo, k_threshold = 0.7, moment_match = FALSE,
   } else {
     model_name <- ""
   }
-  n <- length(loo::pareto_k_ids(loo, threshold = k_threshold))
   ndraws <- dim(loo)[1] %||% Inf
-  if (n > 0 && ndraws < 2200) {
+  n <- n2 <- length(loo::pareto_k_ids(loo, threshold = k_threshold))
+  # for small number of draws the threshold may be smaller than 0.7
+  k_threshold2 <- ps_khat_threshold(ndraws)
+  if (k_threshold2 < k_threshold) {
+    n2 <- length(loo::pareto_k_ids(loo, threshold = k_threshold2))
+  }
+  if (n2 > n && k_threshold2 <= 0.7) {
     warning2(
-      "Found ", n, " observations with a pareto_k > ", k_threshold,
+      "Found ", n2, " observations with a pareto_k > ", round(k_threshold2, 2),
       model_name, ". We recommend to run more iterations to get at least ",
-      "about 2200 posterior draws for more reliable pareteo_k estimation."
+      "about 2200 posterior draws to improve LOO-CV approximation accuracy."
     )
     out <- "loo_more_draws"
   } else if (n > 0 && !moment_match) {
@@ -991,3 +997,9 @@ print.iclist <- function(x, digits = 2, ...) {
   print(round(mat, digits = digits), na.print = "")
   invisible(x)
 }
+
+# Pareto-smoothing k-hat threshold
+# not yet exported by loo so copied over here for now
+ps_khat_threshold <- function(S, ...) {
+  1 - 1 / log10(S)
+}
diff --git a/R/loo_moment_match.R b/R/loo_moment_match.R
index fcf252384..bb998d210 100644
--- a/R/loo_moment_match.R
+++ b/R/loo_moment_match.R
@@ -9,8 +9,8 @@
 #' @inheritParams predict.brmsfit
 #' @param x An object of class \code{brmsfit}.
 #' @param loo An object of class \code{loo} originally created from \code{x}.
-#' @param k_threshold The threshold at which Pareto \eqn{k}
-#'   estimates are treated as problematic. Defaults to \code{0.7}.
+#' @param k_threshold The Pareto \eqn{k} threshold for which observations
+#'   moment matching is applied. Defaults to \code{0.7}.
 #'   See \code{\link[loo:pareto-k-diagnostic]{pareto_k_ids}}
 #'   for more details.
 #' @param check Logical; If \code{TRUE} (the default), some checks
diff --git a/man/loo.brmsfit.Rd b/man/loo.brmsfit.Rd
index 42daf31d5..11ead8904 100644
--- a/man/loo.brmsfit.Rd
+++ b/man/loo.brmsfit.Rd
@@ -53,9 +53,10 @@ details.}
 \item{reloo}{Logical; Indicate whether \code{\link{reloo}}
 should be applied on problematic observations. Defaults to \code{FALSE}.}
 
-\item{k_threshold}{The threshold at which pareto \eqn{k}
-estimates are treated as problematic. Defaults to \code{0.7}.
-Only used if argument \code{reloo} is \code{TRUE}.
+\item{k_threshold}{The Pareto \eqn{k} threshold for which observations
+\code{\link{loo_moment_match}} or \code{\link{reloo}} is applied if
+argument \code{moment_match} or \code{reloo} is \code{TRUE}.
+Defaults to \code{0.7}.
 See \code{\link[loo:pareto-k-diagnostic]{pareto_k_ids}} for more details.}
 
 \item{save_psis}{Should the \code{"psis"} object created internally be saved
diff --git a/man/loo_moment_match.brmsfit.Rd b/man/loo_moment_match.brmsfit.Rd
index 774509882..abab4a19e 100644
--- a/man/loo_moment_match.brmsfit.Rd
+++ b/man/loo_moment_match.brmsfit.Rd
@@ -21,8 +21,8 @@
 
 \item{loo}{An object of class \code{loo} originally created from \code{x}.}
 
-\item{k_threshold}{The threshold at which Pareto \eqn{k}
-estimates are treated as problematic. Defaults to \code{0.7}.
+\item{k_threshold}{The Pareto \eqn{k} threshold for which observations
+moment matching is applied. Defaults to \code{0.7}.
 See \code{\link[loo:pareto-k-diagnostic]{pareto_k_ids}}
 for more details.}