diff --git a/lib/oga/css/lexer.rl b/lib/oga/css/lexer.rl
index ad13ed0..a8b709a 100644
--- a/lib/oga/css/lexer.rl
+++ b/lib/oga/css/lexer.rl
@@ -132,12 +132,12 @@ module Oga
 
         whitespace = [\t ]+;
 
-        comma  = ',' @{ add_token(:T_COMMA) };
-        hash   = '#' @{ add_token(:T_HASH) };
-        dot    = '.' @{ add_token(:T_DOT) };
-        pipe   = '|' @{ add_token(:T_PIPE) };
-        lbrack = '[' @{ add_token(:T_LBRACK) };
-        rbrack = ']' @{ add_token(:T_RBRACK) };
+        comma  = ',' %{ add_token(:T_COMMA) };
+        hash   = '#' %{ add_token(:T_HASH) };
+        dot    = '.' %{ add_token(:T_DOT) };
+        lbrack = '[' %{ add_token(:T_LBRACK) };
+        rbrack = ']' %{ add_token(:T_RBRACK) };
+        pipe   = '|';
 
         # Identifiers
         #
@@ -150,8 +150,49 @@ module Oga
           emit(:T_IDENT, ts, te)
         }
 
+        # Operators
+        #
+        # Various operators that can be used for filtering nodes. For example,
+        # "$=" can be used to select attribute values that end with a given
+        # string.
+        #
+        # http://www.w3.org/TR/css3-selectors/#selectors
+
+        op_eq          = '=';
+        op_space_in    = '~=';
+        op_starts_with = '^=';
+        op_ends_with   = '$=';
+        op_in          = '*=';
+        op_hyphen_in   = '|=';
+        op_child       = '>';
+        op_fol_direct  = '+';
+        op_fol         = '~';
+
         main := |*
-          whitespace | comma | hash | dot | pipe | lbrack | rbrack;
+          whitespace | comma | hash | dot | lbrack | rbrack;
+
+          # Some of the operators have similar characters (e.g. the "="). As a
+          # result we can't use rules like the following:
+          #
+          #     '='  %{ ... };
+          #     '*=' %{ ... };
+          #
+          # This would result in both machines being executed for the input
+          # "*=". The syntax below ensures that only the first match is handled.
+
+          op_eq          => { add_token(:T_EQ) };
+          op_space_in    => { add_token(:T_SPACE_IN) };
+          op_starts_with => { add_token(:T_STARTS_WITH) };
+          op_ends_with   => { add_token(:T_ENDS_WITH) };
+          op_in          => { add_token(:T_IN) };
+          op_hyphen_in   => { add_token(:T_HYPHEN_IN) };
+          op_child       => { add_token(:T_CHILD) };
+          op_fol_direct  => { add_token(:T_FOLLOWING_DIRECT) };
+          op_fol         => { add_token(:T_FOLLOWING) };
+
+          # The pipe character is also used in the |= operator so the action for
+          # this is handled separately.
+          pipe => { add_token(:T_PIPE) };
 
           identifier => emit_identifier;
 
diff --git a/spec/oga/css/lexer/operators_spec.rb b/spec/oga/css/lexer/operators_spec.rb
new file mode 100644
index 0000000..36130bf
--- /dev/null
+++ b/spec/oga/css/lexer/operators_spec.rb
@@ -0,0 +1,41 @@
+require 'spec_helper'
+
+describe Oga::CSS::Lexer do
+  context 'operators' do
+    example 'lex the = operator' do
+      lex_css('=').should == [[:T_EQ, nil]]
+    end
+
+    example 'lex the ~= operator' do
+      lex_css('~=').should == [[:T_SPACE_IN, nil]]
+    end
+
+    example 'lex the ^= operator' do
+      lex_css('^=').should == [[:T_STARTS_WITH, nil]]
+    end
+
+    example 'lex the $= operator' do
+      lex_css('$=').should == [[:T_ENDS_WITH, nil]]
+    end
+
+    example 'lex the *= operator' do
+      lex_css('*=').should == [[:T_IN, nil]]
+    end
+
+    example 'lex the |= operator' do
+      lex_css('|=').should == [[:T_HYPHEN_IN, nil]]
+    end
+
+    example 'lex the > operator' do
+      lex_css('>').should == [[:T_CHILD, nil]]
+    end
+
+    example 'lex the + operator' do
+      lex_css('+').should == [[:T_FOLLOWING_DIRECT, nil]]
+    end
+
+    example 'lex the ~ operator' do
+      lex_css('~').should == [[:T_FOLLOWING, nil]]
+    end
+  end
+end