diff --git a/changelog/unreleased/issue-4583 b/changelog/unreleased/issue-4583
new file mode 100644
index 000000000..97b0e6ba7
--- /dev/null
+++ b/changelog/unreleased/issue-4583
@@ -0,0 +1,12 @@
+Enhancement: Ignore s3.storage-class for metadata if archive tier is specified
+
+There is no official cold storage support in restic, use this option at your
+own risk.
+
+Restic always stored all files on s3 using the specified `s3.storage-class`.
+Now, restic will store metadata using a non-archive storage tier to avoid
+problems when accessing a repository. To restore any data, it is still
+necessary to manually warm up the required data beforehand.
+
+https://github.com/restic/restic/issues/4583
+https://github.com/restic/restic/pull/4584
diff --git a/internal/backend/s3/s3.go b/internal/backend/s3/s3.go
index f0447224f..d41f4479d 100644
--- a/internal/backend/s3/s3.go
+++ b/internal/backend/s3/s3.go
@@ -325,16 +325,29 @@ func (be *Backend) Path() string {
 	return be.cfg.Prefix
 }
 
+// useStorageClass returns whether file should be saved in the provided Storage Class
+// For archive storage classes, only data files are stored using that class; metadata
+// must remain instantly accessible.
+func (be *Backend) useStorageClass(h backend.Handle) bool {
+	notArchiveClass := be.cfg.StorageClass != "GLACIER" && be.cfg.StorageClass != "DEEP_ARCHIVE"
+	isDataFile := h.Type == backend.PackFile && !h.IsMetadata
+	return isDataFile || notArchiveClass
+}
+
 // Save stores data in the backend at the handle.
 func (be *Backend) Save(ctx context.Context, h backend.Handle, rd backend.RewindReader) error {
 	objName := be.Filename(h)
 
-	opts := minio.PutObjectOptions{StorageClass: be.cfg.StorageClass}
-	opts.ContentType = "application/octet-stream"
-	// the only option with the high-level api is to let the library handle the checksum computation
-	opts.SendContentMd5 = true
-	// only use multipart uploads for very large files
-	opts.PartSize = 200 * 1024 * 1024
+	opts := minio.PutObjectOptions{
+		ContentType: "application/octet-stream",
+		// the only option with the high-level api is to let the library handle the checksum computation
+		SendContentMd5: true,
+		// only use multipart uploads for very large files
+		PartSize: 200 * 1024 * 1024,
+	}
+	if be.useStorageClass(h) {
+		opts.StorageClass = be.cfg.StorageClass
+	}
 
 	info, err := be.client.PutObject(ctx, be.cfg.Bucket, objName, io.NopCloser(rd), int64(rd.Length()), opts)