diff --git a/include/cpp-statsd-client/UDPSender.hpp b/include/cpp-statsd-client/UDPSender.hpp
index cdabc75..050c74e 100644
--- a/include/cpp-statsd-client/UDPSender.hpp
+++ b/include/cpp-statsd-client/UDPSender.hpp
@@ -246,8 +246,8 @@ inline void UDPSender::queueMessage(const std::string& message) noexcept {
     // We aquire a lock but only if we actually need to (i.e. there is a thread also accessing the queue)
     auto batchingLock =
         m_batchingThread.joinable() ? std::unique_lock<std::mutex>(m_batchingMutex) : std::unique_lock<std::mutex>();
-    // Either we don't have a place to batch our message or we exceeded the batch size, so make a new batch
-    if (m_batchingMessageQueue.empty() || m_batchingMessageQueue.back().length() > m_batchsize) {
+    // Either we don't have a place to batch our message or we are about to exceed the batch size, so make a new batch
+    if (m_batchingMessageQueue.empty() || m_batchingMessageQueue.back().size() + message.size() > m_batchsize) {
         m_batchingMessageQueue.emplace_back();
         m_batchingMessageQueue.back().reserve(m_batchsize + 256);
     }  // When there is already a batch open we need a separator when its not empty
diff --git a/tests/testStatsdClient.cpp b/tests/testStatsdClient.cpp
index 22e0d6e..aa34101 100644
--- a/tests/testStatsdClient.cpp
+++ b/tests/testStatsdClient.cpp
@@ -9,17 +9,22 @@ using namespace Statsd;
 // Note that we could just synchronously recv metrics and not use a thread but doing the test async has the
 // advantage that we can test the threaded batching mode in a straightforward way. The server thread basically
 // just keeps storing metrics in an vector until it hears a special one signaling the test is over and bails
-void mock(StatsdServer& server, std::vector<std::string>& messages) {
+void mock(StatsdServer& server, std::vector<std::string>& messages, bool split) {
     do {
         // Grab the messages that are waiting
         auto recvd = server.receive();
 
+        // If we don't want splitting, give the raw message though we still go
+        // through splitting to parse the quit message
+        if (!split) messages.emplace_back(recvd);
+
         // Split the messages on '\n'
         auto start = std::string::npos;
         do {
-            // Keep this message
+            // Keep this message (unless we don't want splitting, then we
+            // do nothing because we already stored the whole batch)
             auto end = recvd.find('\n', ++start);
-            messages.emplace_back(recvd.substr(start, end));
+            if (split) messages.emplace_back(recvd.substr(start, end));
             start = end;
 
             // Bail if we found the special quit message
@@ -53,6 +58,72 @@ void testErrorConditions() {
     throwOnError(client, false, "Should not be able to resolve a ridiculous ip");
 }
 
+void testOverflow() {
+    StatsdServer mock_server;
+    std::vector<std::string> messages, expected;
+    std::thread server(mock, std::ref(mock_server), std::ref(messages), /*split=*/false);
+
+    // Set a new config that has the client send messages to a proper address that can be resolved
+    StatsdClient client("localhost", 8125, /*prefix=*/"", /*batchsize=*/48, /*sendInterval=*/0, /*gaugePrecision=*/3);
+    throwOnError(client);
+
+    // Send message that exhausts most of the batch limit (40/48)
+    client.increment("stat.timesWeUsed40CharsOfBufferSpace");
+    throwOnError(client);
+    expected.emplace_back("stat.timesWeUsed40CharsOfBufferSpace:1|c");
+
+    // Send message that would go over the batch limit (50/48)
+    // The second message shouldn't fit the batch and must be part of the next batch
+    client.increment("stat.i");
+    throwOnError(client);
+    expected.emplace_back("stat.i:1|c");
+
+    // Flush messages
+    client.flush();
+
+    // Send messages that would exactly fit in the batch, newlines included (48/48)
+    // Everything should be part of the same batch
+    expected.emplace_back();
+    for (int idx = 0; idx < 5; idx++) {
+        client.increment("silly");
+        throwOnError(client);
+        expected.back().append("silly:1|c\n");
+    }
+    expected.back().pop_back();
+
+    // Flush messages
+    client.flush();
+
+    // With all messages flushed, send one giant message that would exceed the batch (50/48)
+    // The message should arrive intact in one batch
+    client.increment("stat.veryImportantIfIncrementedComputerExplode");
+    throwOnError(client);
+    expected.emplace_back("stat.veryImportantIfIncrementedComputerExplode:1|c");
+
+    // Flush messages
+    client.flush();
+
+    // Signal the mock server we are done (and flush so that the signal is sent)
+    client.timing("DONE", 0);
+    client.flush();
+
+    // Wait for the server to stop
+    server.join();
+
+    // Make sure we get the exactly correct output
+    if (messages != expected) {
+        std::cerr << "Unexpected stats received by server, got:" << std::endl;
+        for (const auto& message : messages) {
+            std::cerr << "- " << message << std::endl;
+        }
+        std::cerr << std::endl << "But we expected:" << std::endl;
+        for (const auto& message : expected) {
+            std::cerr << "- " << message << std::endl;
+        }
+        throw std::runtime_error("Unexpected stats");
+    }
+}
+
 void testReconfigure() {
     StatsdServer server;
     throwOnError(server);
@@ -79,7 +150,7 @@ void testReconfigure() {
 void testSendRecv(uint64_t batchSize, uint64_t sendInterval) {
     StatsdServer mock_server;
     std::vector<std::string> messages, expected;
-    std::thread server(mock, std::ref(mock_server), std::ref(messages));
+    std::thread server(mock, std::ref(mock_server), std::ref(messages), /*split=*/true);
 
     // Set a new config that has the client send messages to a proper address that can be resolved
     StatsdClient client("localhost", 8125, "sendRecv.", batchSize, sendInterval, 3);
@@ -176,6 +247,8 @@ int main() {
 
     // general things that should be errors
     testErrorConditions();
+    // ensure batch size is respected
+    testOverflow();
     // reconfiguring how you are sending
     testReconfigure();
     // no batching