diff --git a/R/getMDS.R b/R/getMDS.R
index d1c200a51..2c9b17de6 100644
--- a/R/getMDS.R
+++ b/R/getMDS.R
@@ -17,6 +17,14 @@
 #' difference is that these functions apply microbiome-specific options such
 #' as \code{getDissimilarity} function by default.
 #'
+#' By default mia does not filter features (sets \code{ntop = Inf}), so all
+#' taxa are retained before running MDS. Note that
+#' \code{scater::calculateMDS} itself uses \code{ntop = 500} by default.
+#' When rarefaction is enabled (\code{niter > 0}) the sampling depth is
+#' evaluated on the retained matrix, so samples whose depth falls below the
+#' requested \code{sample} are dropped. Use \code{subset.row} to explicitly
+#' subset rows if desired.
+#'
 #' See \code{\link[scater:runMDS]{scater::calculateMDS}} for details.
 #'
 #' @return
@@ -40,6 +48,14 @@
 #'     \item \code{subset.result}: \code{Logical result}. Specifies whether to
 #'     subset \code{x} to match the result if some samples were removed during
 #'     calculation. (Default: \code{TRUE})
+#'
+#'     \item \code{ntop}: \code{Numeric scalar}. Forwarded to
+#'     \code{scater::calculateMDS}. Controls how many of the most variable
+#'     features are kept before MDS is run. (Default: \code{Inf}). Set
+#'     a finite \code{ntop} (or supply \code{subset.row}) when you want to
+#'     filter features before running MDS; use \code{ntop = Inf} when the full
+#'     assay is required (for example, for rarefaction where absolute library
+#'     sizes matter).
 #' }
 #'
 #' @examples
@@ -60,6 +76,18 @@
 #' # Calculate PCoA with Unifrac and rarefaction. (Note: increase iterations)
 #' tse <- addMDS(tse, method = "unifrac", name = "unifrac_rare", niter = 2L)
 #'
+#' # Disable feature filtering when rarefying at the minimum depth
+#' min_depth <- min(colSums(assay(tse, "counts")))
+#' tse <- addMDS(
+#'     tse,
+#'     assay.type = "counts",
+#'     method = "bray",
+#'     niter = 2L,
+#'     sample = min_depth,
+#'     ntop = Inf
+#' )
+#' reducedDim(tse, "MDS")[1:3, 1:2]
+#'
 #' # Visualize results
 #' p1 <- plotReducedDim(tse, "unifrac", colour_by = "SampleType") +
 #'     labs(title = "Not rarefied")
@@ -122,6 +150,11 @@ setMethod("getMDS", signature = c(x = "TreeSummarizedExperiment"),
 # This function is used to set default options for SCE
 .get_mds_args <- function(x, assay.type, FUN = getDissimilarity, ...){
     args <- c(list(x = x, assay.type = assay.type, FUN = FUN), list(...))
+    # Default to no filtering (retain all features) unless user provided ntop
+    arg_names <- names(args)
+    if( is.null(arg_names) || !("ntop" %in% arg_names) ){
+        args$ntop <- Inf
+    }
     return(args)
 }
 
diff --git a/man/addMDS.Rd b/man/addMDS.Rd
index 85e3b7072..680850505 100644
--- a/man/addMDS.Rd
+++ b/man/addMDS.Rd
@@ -29,6 +29,14 @@ calculate dissimilarity. (Default: \code{getDissimilarity})
 \if{html}{\out{<div class="sourceCode">}}\preformatted{\\item \code{subset.result}: \code{Logical result}. Specifies whether to
 subset \code{x} to match the result if some samples were removed during
 calculation. (Default: \code{TRUE})
+
+\\item \code{ntop}: \code{Numeric scalar}. Forwarded to
+\code{scater::calculateMDS}. Controls how many of the most variable
+features are kept before MDS is run. (Default: \code{Inf}). Set
+a finite \code{ntop} (or supply \code{subset.row}) when you want to
+filter features before running MDS; use \code{ntop = Inf} when the full
+assay is required (for example, for rarefaction where absolute library
+sizes matter).
 }\if{html}{\out{</div>}}
 
 }}
@@ -57,6 +65,14 @@ returns the results, \code{addMDS} adds them to \code{reducedDim(x)}. The
 difference is that these functions apply microbiome-specific options such
 as \code{getDissimilarity} function by default.
 
+By default mia does not filter features (sets \code{ntop = Inf}), so all
+taxa are retained before running MDS. Note that
+\code{scater::calculateMDS} itself uses \code{ntop = 500} by default.
+When rarefaction is enabled (\code{niter > 0}) the sampling depth is
+evaluated on the retained matrix, so samples whose depth falls below the
+requested \code{sample} are dropped. Use \code{subset.row} to explicitly
+subset rows if desired.
+
 See \code{\link[scater:runMDS]{scater::calculateMDS}} for details.
 }
 \examples{
@@ -77,6 +93,18 @@ tse <- addMDS(tse, method = "unifrac", name = "unifrac")
 # Calculate PCoA with Unifrac and rarefaction. (Note: increase iterations)
 tse <- addMDS(tse, method = "unifrac", name = "unifrac_rare", niter = 2L)
 
+# Disable feature filtering when rarefying at the minimum depth
+min_depth <- min(colSums(assay(tse, "counts")))
+tse <- addMDS(
+    tse,
+    assay.type = "counts",
+    method = "bray",
+    niter = 2L,
+    sample = min_depth,
+    ntop = Inf
+)
+reducedDim(tse, "MDS")[1:3, 1:2]
+
 # Visualize results
 p1 <- plotReducedDim(tse, "unifrac", colour_by = "SampleType") +
     labs(title = "Not rarefied")
diff --git a/tests/testthat/test-addMDS.R b/tests/testthat/test-addMDS.R
index 2b7df7ad9..79ab4ed46 100644
--- a/tests/testthat/test-addMDS.R
+++ b/tests/testthat/test-addMDS.R
@@ -14,15 +14,15 @@ test_that("Compare addMDS with runMDS", {
     tse <- GlobalPatterns
     #
     res <- getMDS(tse, assay.type = "counts", method = "bray")
-    ref <- calculateMDS(tse, assay.type = "counts", method = "bray", FUN = getDissimilarity)
+    ref <- calculateMDS(tse, assay.type = "counts", method = "bray", FUN = getDissimilarity, ntop = Inf)
     expect_equal(res, ref)
     #
     res <- addMDS(tse, assay.type = "counts", method = "bray", name = "test")
-    ref <- runMDS(tse, assay.type = "counts", method = "bray", FUN = getDissimilarity, name = "test")
+    ref <- runMDS(tse, assay.type = "counts", method = "bray", FUN = getDissimilarity, ntop = Inf, name = "test")
     expect_equal(reducedDim(res, "test"), reducedDim(ref, "test"))
     #
-    res <- getMDS(tse, assay.type = "counts", method = "unifrac", name = "test") |> expect_warning()
-    ref <- calculateMDS(tse, assay.type = "counts", method = "unifrac", tree = rowTree(tse), FUN = getDissimilarity, name = "test") |> expect_warning()
+    res <- getMDS(tse, assay.type = "counts", method = "unifrac", ntop = 500, name = "test") |> expect_warning()
+    ref <- calculateMDS(tse, assay.type = "counts", method = "unifrac", tree = rowTree(tse), FUN = getDissimilarity, ntop = 500, name = "test") |> expect_warning()
     expect_equal(res, ref)
 })