diff --git a/Test.hs b/Test.hs
index d9fdca9951..8e16f50b2e 100644
--- a/Test.hs
+++ b/Test.hs
@@ -1738,6 +1738,11 @@ test_export_import = intmpclonerepoInDirect $ do
 	git_annex "get" [] @? "get of files failed"
 	annexed_present annexedfile
 
+	-- When on an adjusted branch, this updates the master branch
+	-- to match it, which is necessary since the master branch is going
+	-- to be exported.
+	git_annex "sync" [] @? "sync failed"
+
 	git_annex "export" ["master", "--to", "foo"] @? "export to dir failed"
 	dircontains annexedfile (content annexedfile)
 
@@ -1793,6 +1798,11 @@ test_export_import_subdir = intmpclonerepoInDirect $ do
 		@? "git mv failed"
 	boolSystem "git" [Param "commit", Param "-m", Param "moved"]
 		@? "git commit failed"
+	
+	-- When on an adjusted branch, this updates the master branch
+	-- to match it, which is necessary since the master branch is going
+	-- to be exported.
+	git_annex "sync" [] @? "sync failed"
 
 	-- Run three times because there was a bug that took a couple
 	-- of runs to lead to the wrong tree being written to the remote
diff --git a/doc/bugs/windows_export_tree_exports_empty_tree.mdwn b/doc/bugs/windows_export_tree_exports_empty_tree.mdwn
new file mode 100644
index 0000000000..45af8f600c
--- /dev/null
+++ b/doc/bugs/windows_export_tree_exports_empty_tree.mdwn
@@ -0,0 +1,24 @@
+The test suite's `test_export_import` fails on Windows.
+It seems that exporting a tree that contains annexed files
+somehow ends up exporting an empty tree.
+
+The test is running in an adjusted unlocked branch. But it exports
+the master branch. The master branch indeed contains an empty tree.
+
+The origin repo is also using an adjusted unlocked branch. Some changes
+have been committed there, but not synced back to the master branch.
+
+On clone from that, git-annex sets up an adjusted unlocked branch,
+and it merges origin's adjusted unlocked branch in. So the annexed files
+are present in that branch, but still not in master.
+
+This is not really a windows-specific problem. Only on windows or perhaps
+on a crippled FS does the test suite actually test opeation in an unlocked
+adjusted branch, because it enters that mode only on demand. Arguably
+the test suite should run all the tests separately on an adjusted branch,
+but that would add a lot of extra time to the test suite and very few
+tests are likely to behave any differently there.
+
+On balance, I think that making the test case run a git-annex sync
+before exporting is good enough.
+--[[Joey]]