diff --git a/docs/bitset.adoc b/docs/bitset.adoc
index 57c72a2..de28fe0 100644
--- a/docs/bitset.adoc
+++ b/docs/bitset.adoc
@@ -101,6 +101,13 @@ auto bs = stdx::bitset<8>{0b1010'1010ul};
 for_each([&](auto i) { /* i == 1, 3, 5, 7 */ }, bs);
 ----
 
+To iterate all the unset bits, use:
+[source,cpp]
+----
+auto bs = stdx::bitset<8>{0b1010'1010ul};
+for_each<stdx::unset_bit>([&](auto i) { /* i == 0, 2, 4, 6 */ }, bs);
+----
+
 To support "external" iteration, or use cases like using a bitset to track used
 objects, `lowest_unset` is also provided:
 [source,cpp]
diff --git a/include/stdx/bitset.hpp b/include/stdx/bitset.hpp
index 36e1187..a48b7d9 100644
--- a/include/stdx/bitset.hpp
+++ b/include/stdx/bitset.hpp
@@ -21,6 +21,45 @@
 
 namespace stdx {
 inline namespace v1 {
+struct set_bit {
+    template <auto Bit, typename IterArg, auto Mask>
+    constexpr static auto fn(auto e, auto idx, auto &f) {
+        using elem_t = decltype(Bit);
+        while (e != 0) {
+            auto const offset = static_cast<std::size_t>(countr_zero(e));
+            e &= static_cast<elem_t>(~(Bit << offset));
+            f(static_cast<IterArg>(idx + offset));
+        }
+    }
+};
+struct unset_bit {
+    template <auto Bit, typename IterArg, auto Mask>
+    constexpr static auto fn(auto e, auto idx, auto &f) {
+        using elem_t = decltype(Bit);
+        while (e != Mask) {
+            auto const offset = static_cast<std::size_t>(countr_one(e));
+            e |= static_cast<elem_t>(Bit << offset);
+            f(static_cast<IterArg>(idx + offset));
+        }
+    }
+};
+struct bit {
+    template <auto Bit, typename IterArg, auto Mask>
+    constexpr static auto fn(auto e, auto idx, auto &f) {
+        using elem_t = decltype(Bit);
+        for (auto i = std::size_t{}; i < popcount(Mask); ++i) {
+            bool b = e & static_cast<elem_t>(Bit << i);
+            f(static_cast<IterArg>(idx + i), b);
+        }
+    }
+};
+
+namespace detail {
+template <typename T>
+concept bit_spec = std::same_as<T, set_bit> or std::same_as<T, unset_bit> or
+                   std::same_as<T, bit>;
+}
+
 template <auto Size,
           typename StorageElem = decltype(smallest_uint<to_underlying(Size)>())>
 class bitset {
@@ -106,20 +145,20 @@ class bitset {
         return not std::is_enum_v<T> or std::is_same_v<T, decltype(Size)>;
     }
 
-    template <typename F> constexpr auto for_each(F &&f) const -> F {
-        std::size_t i = 0;
-        for (auto e : storage) {
-            while (e != 0) {
-                auto const offset = static_cast<std::size_t>(countr_zero(e));
-                e &= static_cast<elem_t>(~(bit << offset));
-                f(static_cast<iter_arg_t>(i + offset));
-            }
-            i += std::numeric_limits<elem_t>::digits;
+    template <detail::bit_spec Spec, typename F>
+    constexpr auto for_each(F &&f) const -> F {
+        std::size_t idx = 0;
+        for (auto i = std::size_t{}; i < storage_size - 1; ++i) {
+            Spec::template fn<bit, iter_arg_t,
+                              std::numeric_limits<elem_t>::max()>(storage[i],
+                                                                  idx, f);
+            idx += std::numeric_limits<elem_t>::digits;
         }
+        Spec::template fn<bit, iter_arg_t, lastmask>(highbits(), idx, f);
         return std::forward<F>(f);
     }
 
-    template <typename F, auto M, typename... S>
+    template <detail::bit_spec Spec, typename F, auto M, typename... S>
     friend constexpr auto for_each(F &&f, bitset<M, S> const &...bs) -> F;
 
     template <typename T, typename F, typename R>
@@ -457,10 +496,10 @@ class bitset {
     }
 };
 
-template <typename F, auto M, typename... S>
+template <detail::bit_spec Spec = set_bit, typename F, auto M, typename... S>
 constexpr auto for_each(F &&f, bitset<M, S> const &...bs) -> F {
     if constexpr (sizeof...(bs) == 1) {
-        return (bs.for_each(std::forward<F>(f)), ...);
+        return (bs.template for_each<Spec>(std::forward<F>(f)), ...);
     } else {
         static_assert(stdx::always_false_v<F>, "unimplemented");
         return f;
@@ -648,9 +687,10 @@ template <typename... Ts> class type_bitset {
         return *this;
     }
 
-    template <typename F> constexpr auto for_each(F &&f) const -> F {
+    template <detail::bit_spec Spec = set_bit, typename F>
+    constexpr auto for_each(F &&f) const -> F {
         constexpr auto callers = make_callers<F>(std::make_index_sequence<N>{});
-        stdx::for_each([&](auto i) { callers[i](f); }, bs);
+        stdx::for_each<Spec>([&](auto i) { callers[i](f); }, bs);
         return f;
     }
 };
diff --git a/test/bitset.cpp b/test/bitset.cpp
index 97557ce..ebe13c2 100644
--- a/test/bitset.cpp
+++ b/test/bitset.cpp
@@ -350,6 +350,23 @@ TEMPLATE_TEST_CASE("for_each", "[bitset]", std::uint8_t, std::uint16_t,
     CHECK(result == bs);
 }
 
+TEMPLATE_TEST_CASE("for_each (unset bits)", "[bitset]", std::uint8_t,
+                   std::uint16_t, std::uint32_t, std::uint64_t) {
+    constexpr auto bs = stdx::bitset<64, TestType>{0x01020304'05060708ul};
+    auto result = decltype(bs){};
+    for_each<stdx::unset_bit>([&](auto i) { result.set(i); }, bs);
+    CHECK(result.template to<std::uint64_t>() ==
+          ~bs.template to<std::uint64_t>());
+}
+
+TEMPLATE_TEST_CASE("for_each (all bits)", "[bitset]", std::uint8_t,
+                   std::uint16_t, std::uint32_t, std::uint64_t) {
+    constexpr auto bs = stdx::bitset<64, TestType>{0x01020304'05060708ul};
+    auto result = decltype(bs){};
+    for_each<stdx::bit>([&](auto i, bool b) { result.set(i, b); }, bs);
+    CHECK(result == bs);
+}
+
 TEMPLATE_TEST_CASE("for_each iterates in order lsb to msb", "[bitset]",
                    std::uint8_t, std::uint16_t, std::uint32_t, std::uint64_t) {
     constexpr auto bs = stdx::bitset<5, TestType>{0b10101ul};
diff --git a/test/type_bitset.cpp b/test/type_bitset.cpp
index a899474..d20c76d 100644
--- a/test/type_bitset.cpp
+++ b/test/type_bitset.cpp
@@ -200,3 +200,12 @@ TEST_CASE("for_each", "[type_bitset]") {
     });
     CHECK(result == "int0float1bool2");
 }
+
+TEST_CASE("for_each (unset bits)", "[type_bitset]") {
+    constexpr auto bs = stdx::type_bitset<int, float, bool>{};
+    auto result = std::string{};
+    bs.for_each<stdx::unset_bit>([&]<typename T, std::size_t I>() -> void {
+        result += std::string{stdx::type_as_string<T>()} + std::to_string(I);
+    });
+    CHECK(result == "int0float1bool2");
+}