diff --git a/m-string.h b/m-string.h
index fe7c2eb8..4667dd30 100644
--- a/m-string.h
+++ b/m-string.h
@@ -657,6 +657,7 @@ string_fget_word (string_t v, const char separator[], FILE *f)
 {
   char buffer[128];
   char c = 0;
+  int d;
   STRINGI_CONTRACT(v);
   assert (f != NULL);
   assert (1+20+2+strlen(separator)+3 < sizeof buffer);
@@ -664,6 +665,14 @@ string_fget_word (string_t v, const char separator[], FILE *f)
   v->size = 0;
   v->ptr[0] = 0;
   bool retcode = false;
+  /* Skip separator first */
+  do {
+          d = fgetc(f);
+          if (d == EOF) {
+            return false;
+          }
+  } while (strchr(separator, d) != NULL);
+  ungetc(d, f);
   /* NOTE: We generate a buffer which we give to scanf to parse the string,
      that it is to say, we generate the format dynamically!
      The format is like " %49[^ \t.\n]%c"
diff --git a/tests/test-mstring.c b/tests/test-mstring.c
index 6f0e0a70..0501e4f5 100644
--- a/tests/test-mstring.c
+++ b/tests/test-mstring.c
@@ -409,7 +409,7 @@ static void test0(void)
   string_init(s1);
   f = fopen("a-mstring.dat", "wt");
   assert(f != NULL);
-  fprintf(f, "hello world.\n\tHowwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwww do you do?");
+  fprintf(f, "hello world...\n\tHowwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwww do you do?");
   fclose(f);
   f = fopen("a-mstring.dat", "rt");
   b = string_fget_word(s1, " \t.\n?", f);