diff --git a/.travis.yml b/.travis.yml
index 8f11202..f9d66a0 100644
--- a/.travis.yml
+++ b/.travis.yml
@@ -7,6 +7,9 @@ sudo: false
 script:
   - cargo build --verbose
   - cargo test --verbose
+  - cargo package
+  - cd target/package/unicode-normalization-*
+  - RUSTFLAGS="--cfg minimal_tests" cargo test --verbose
 notifications:
   email:
     on_success: never
diff --git a/Cargo.toml b/Cargo.toml
index e4727c7..9c82408 100644
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -18,7 +18,7 @@ Decomposition and Recomposition, as described in
 Unicode Standard Annex #15.
 """
 
-exclude = [ "target/*", "Cargo.lock", "scripts/tmp", "*.txt", "src/normalization_tests.rs", "src/test.rs" ]
+exclude = [ "target/*", "Cargo.lock", "scripts/tmp", "*.txt", "src/normalization_tests.rs" ]
 
 [dependencies]
 smallvec = "0.6"
\ No newline at end of file
diff --git a/README.md b/README.md
index 591849d..d8d4fb3 100644
--- a/README.md
+++ b/README.md
@@ -33,3 +33,26 @@ to your `Cargo.toml`:
 [dependencies]
 unicode-normalization = "0.1.8"
 ```
+
+## Linux Vendors / Downstream
+As is, tests won't work on the published crate, as important
+corpus data required for fully testing functionality otherwise
+bloats the size of the crate.
+
+Tests aren't hugely meaningful without this, but there are two
+workarounds:
+
+```bash
+RUSTFLAGS="--cfg minimal_tests" cargo test
+```
+
+This will make the crate compile, and some arbitrary set of lower
+quality tests pass.
+
+```bash
+python scripts/unicode.py
+cp ./normalization_tests.rs src/
+```
+
+This will generate the full corpus required for tests to work,
+without needing to pass any special flags.
diff --git a/src/lib.rs b/src/lib.rs
index 7a4b4e0..9aebcbb 100644
--- a/src/lib.rs
+++ b/src/lib.rs
@@ -75,7 +75,7 @@ mod tables;
 
 #[cfg(test)]
 mod test;
-#[cfg(test)]
+#[cfg(all(test,not(minimal_tests)))]
 mod normalization_tests;
 
 /// Methods for composing and decomposing characters.
diff --git a/src/stream_safe.rs b/src/stream_safe.rs
index 38bb42c..25a7ae1 100644
--- a/src/stream_safe.rs
+++ b/src/stream_safe.rs
@@ -111,6 +111,7 @@ mod tests {
         classify_nonstarters,
     };
     use std::char;
+    #[cfg(not(minimal_tests))]
     use normalization_tests::NORMALIZATION_TESTS;
     use normalize::decompose_compatible;
     use lookups::canonical_combining_class;
@@ -119,6 +120,7 @@ mod tests {
         StreamSafe::new(s.chars()).collect()
     }
 
+    #[cfg(not(minimal_tests))]
     #[test]
     fn test_normalization_tests_unaffected() {
         for test in NORMALIZATION_TESTS {
diff --git a/src/test.rs b/src/test.rs
index 4c7d2eb..c0864ce 100644
--- a/src/test.rs
+++ b/src/test.rs
@@ -95,6 +95,7 @@ fn test_nfkc() {
     t!("a\u{300}\u{305}\u{315}\u{5ae}b", "\u{e0}\u{5ae}\u{305}\u{315}b");
 }
 
+#[cfg(not(minimal_tests))]
 #[test]
 fn test_official() {
     use normalization_tests::NORMALIZATION_TESTS;
@@ -158,7 +159,7 @@ fn test_official() {
         }
     }
 }
-
+#[cfg(not(minimal_tests))]
 #[test]
 fn test_quick_check() {
     use normalization_tests::NORMALIZATION_TESTS;