diff --git a/ChangeLog b/ChangeLog
index 966028a86..4186f0fcc 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,4 +1,11 @@
-2004-06-23  Alexandre Duret-Lutz  <adl@gnu.org>
+2004-06-23  Alexandre Duret-Lutz  <adl@src.lip6.fr>
+
+	* src/ltlvisit/tostring.cc (is_bare_word): New function.
+	(to_string_visitor::visitor(const atomic_prop*)): Use is_bare_word
+	to better check which atomic proposition need to be quoted.
+	* src/ltlparse/ltlscan.ll: Do not allow identifiers starting with F_
+	or G_.
+	* src/ltltest/equals.test, src/ltltest/tostring.test: More tests.
 
 	* src/tgba/tgbatba.cc (tgba_tba_proxy::format_state): Use
 	bdd_format_accset to print the acceptance condition part of the state.
diff --git a/src/ltlparse/ltlscan.ll b/src/ltlparse/ltlscan.ll
index d7f020da5..906f2579c 100644
--- a/src/ltlparse/ltlscan.ll
+++ b/src/ltlparse/ltlscan.ll
@@ -96,10 +96,10 @@ flex_set_buffer(const char* buf)
      used by a unary operator (F,G,X), unless this
      letter is followed by a digit in which case we assume
      it's an ATOMIC_PROP (even though F0 could be seen as Ffalse, we
-     don't).
+     don't, because Ffalse is never used in practice).
   */
 <INITIAL>[a-zA-EH-WYZ_][a-zA-Z0-9_]* |
-<INITIAL>[FGX][0-9_][a-zA-Z0-9_]* |
+<INITIAL>[FGX][0-9][a-zA-Z0-9_]* |
   /*
      However if we have just parsed an atomic proposition, then we
      are not expecting another atomic proposition, so we can be stricter
diff --git a/src/ltltest/equals.test b/src/ltltest/equals.test
index 858bb60fe..80e040b27 100755
--- a/src/ltltest/equals.test
+++ b/src/ltltest/equals.test
@@ -57,3 +57,6 @@ run 1 ./equals 'a & b & (c |(f U g)| e)' \
 
 # Precedence
 run 0 ./equals 'a & b ^ c | d' 'd | c ^ b & a'
+
+# Corner cases parsing
+run 0 ./equals 'FFG__GFF' 'F(F(G("__GFF")))'
diff --git a/src/ltltest/tostring.test b/src/ltltest/tostring.test
index ed1828808..453ad9c12 100755
--- a/src/ltltest/tostring.test
+++ b/src/ltltest/tostring.test
@@ -51,3 +51,6 @@ run 0 ./tostring 'G"TruE"'
 run 0 ./tostring 'FFALSE'
 run 0 ./tostring 'GTruE'
 run 0 ./tostring 'p=0UFXp=1'
+run 0 ./tostring 'GF"\GF"'
+run 0 ./tostring 'GF"foo bar"'
+run 0 ./tostring 'FFG__GFF'
diff --git a/src/ltlvisit/tostring.cc b/src/ltlvisit/tostring.cc
index 9f6d6e3db..876f1290f 100644
--- a/src/ltlvisit/tostring.cc
+++ b/src/ltlvisit/tostring.cc
@@ -21,6 +21,7 @@
 
 #include <cassert>
 #include <sstream>
+#include <ctype.h>
 #include "tostring.hh"
 #include "ltlast/visitor.hh"
 #include "ltlast/allnodes.hh"
@@ -31,6 +32,27 @@ namespace spot
   namespace ltl
   {
 
+    static bool
+    is_bare_word(const char* str)
+    {
+      // Bare words cannot be empty, start with the letter of a unary
+      // operator, or be the name of an existing constant.  Also they
+      // should start with an letter.
+      if (!*str
+	  || *str == 'F'
+	  || *str == 'G'
+	  || *str == 'X'
+	  || !(isalpha(*str) || *str == '_')
+	  || !strcasecmp(str, "true")
+	  || !strcasecmp(str, "false"))
+	return false;
+      // The remaining of the word must be alphanumeric.
+      while (*++str)
+	if (!(isalnum(*str) || *str == '_'))
+	  return false;
+      return true;
+    }
+
     class to_string_visitor : public const_visitor
     {
     public:
@@ -48,9 +70,7 @@ namespace spot
       visit(const atomic_prop* ap)
       {
 	std::string str = ap->name();
-	if (str[0] == 'F' || str[0] == 'G' || str[0] == 'X'
-	    || !strcasecmp(str.c_str(), "true")
-	    || !strcasecmp(str.c_str(), "false"))
+	if (!is_bare_word(str.c_str()))
 	  {
 	    os_ << '"' << str << '"';
 	  }