diff --git a/base/base_tests/string_utils_test.cpp b/base/base_tests/string_utils_test.cpp
index 49720cc45b..bac640f4d3 100644
--- a/base/base_tests/string_utils_test.cpp
+++ b/base/base_tests/string_utils_test.cpp
@@ -733,13 +733,16 @@ UNIT_TEST(NormalizeDigits_UniString)
 UNIT_TEST(Split)
 {
   vector<string> target;
-  strings::Split(";Test\\;проверка;0;", ';', target);
+  TEST(strings::ParseCSVRow(",Test\\,проверка,0,", target), ());
   vector<string> expected({"", "Test\\", "проверка", "0", ""});
   TEST_EQUAL(target, expected, ());
-  strings::Split("and there  was none", ' ', target);
-  vector<string> expected2({"and", "there", "", "was", "none"});
+  TEST(strings::ParseCSVRow("and there  \"was  none\"", target, ' '), ());
+  vector<string> expected2({"and", "there", "", "was  none"});
   TEST_EQUAL(target, expected2, ());
-  strings::Split("", '!', target);
+  TEST(!strings::ParseCSVRow("", target), ());
   vector<string> expected3;
   TEST_EQUAL(target, expected3, ());
+  TEST(!strings::ParseCSVRow("\"this, a line.\"", target, ',', 2), (target));
+  vector<string> expected4({"this, a line."});
+  TEST_EQUAL(target, expected4, ());
 }
diff --git a/base/string_utils.cpp b/base/string_utils.cpp
index 2a8b029f37..ab53723635 100644
--- a/base/string_utils.cpp
+++ b/base/string_utils.cpp
@@ -328,17 +328,61 @@ bool AlmostEqual(string const & str1, string const & str2, size_t mismatchedCoun
   return false;
 }
 
-void Split(string const & s, char delimiter, vector<string> & target)
+bool ParseCSVRow(string const & s, vector<string> & target, char const delimiter, size_t const columns)
 {
   target.clear();
+  using It = TokenizeIterator<SimpleDelimiter, string::const_iterator, true>;
+  bool insideQuotes = false;
+  ostringstream quoted;
+  for (It it(s, SimpleDelimiter(delimiter)); it; ++it)
+  {
+    string column = *it;
+    if (insideQuotes)
+    {
+      if (!column.empty() && column.back() == '"')
+      {
+        // Found the tail quote: remove it and add |quoted| to the vector.
+        insideQuotes = false;
+        column.pop_back();
+        quoted << delimiter << column;
+        target.push_back(quoted.str());
+        quoted.clear();
+      }
+      else
+        quoted << delimiter << column;
+    }
+    else if (!column.empty() && column.front() == '"')
+    {
+      // Found the front quote: if there is the last one also, remove both and append column,
+      // otherwise push the column into a |quoted| buffer.
+      column.erase(0, 1);
+      if (column.back() == '"')
+      {
+        column.pop_back();
+        strings::Trim(column);
+        target.push_back(column);
+      }
+      else
+      {
+        quoted << column;
+        insideQuotes = true;
+      }
+    }
+    else
+    {
+      strings::Trim(column);
+      target.push_back(column);
+    }
+  }
 
   // Special case: if the string is empty, return an empty array instead of {""}.
-  if (s.empty())
-    return;
+  if (target.size() == 1 && target[0].empty())
+  {
+    target.clear();
+    return false;
+  }
 
-  using It = TokenizeIterator<SimpleDelimiter, string::const_iterator, true>;
-  for (It it(s, SimpleDelimiter(delimiter)); it; ++it)
-    target.push_back(*it);
+  return columns <= 0 || target.size() == columns;
 }
 
 }  // namespace strings
diff --git a/base/string_utils.hpp b/base/string_utils.hpp
index e42c8188e8..f9883a544c 100644
--- a/base/string_utils.hpp
+++ b/base/string_utils.hpp
@@ -307,7 +307,9 @@ void Tokenize(string const & str, char const * delims, TFunctor && f)
 }
 
 /// Splits a string by the delimiter, keeps empty parts, on an empty string returns an empty vector.
-void Split(string const & s, char delimiter, vector<string> & target);
+/// Supports quoted columns, does not support newlines in columns and escaped quotes.
+/// @return false if the line is empty or number of columns differs from |columns|.
+bool ParseCSVRow(string const & s, vector<string> & target, char const delimiter = ',', size_t const columns = 0);
 
 /// @return code of last symbol in string or 0 if s is empty
 UniChar LastUniChar(string const & s);
diff --git a/generator/booking_dataset.cpp b/generator/booking_dataset.cpp
index 2e619453b5..325beaee3c 100644
--- a/generator/booking_dataset.cpp
+++ b/generator/booking_dataset.cpp
@@ -30,10 +30,8 @@ bool CheckForValues(string const & value)
 
 BookingDataset::Hotel::Hotel(string const & src)
 {
-  vector<string> rec(FieldsCount());
-  strings::SimpleTokenizer token(src, "\t");
-  for (size_t i = 0; token && i < rec.size(); ++i, ++token)
-    rec[i] = *token;
+  vector<string> rec;
+  CHECK(strings::ParseCSVRow(src, rec, '\t', FieldsCount()), ("Error parsing hotels.tsv line:", src));
 
   strings::to_uint(rec[Index(Fields::Id)], id);
   strings::to_double(rec[Index(Fields::Latitude)], lat);
@@ -176,7 +174,7 @@ void BookingDataset::BuildFeatures(function<void(OsmElement *)> const & fn) cons
     if (!hotel.translations.empty())
     {
       vector<string> parts;
-      strings::Split(hotel.translations, '|', parts);
+      strings::ParseCSVRow(hotel.translations, parts, '|');
       for (auto i = 0; i < parts.size(); i += 3)
       {
         e.AddTag("name:" + parts[i], parts[i + 1]);