diff --git a/src/p_lx_elf.cpp b/src/p_lx_elf.cpp
index 6f4407ab..5cdb7742 100644
--- a/src/p_lx_elf.cpp
+++ b/src/p_lx_elf.cpp
@@ -738,14 +738,7 @@ void PackLinuxElf32::pack2(OutputFile *fo, Filter &ft)
     unsigned total_in = 0;
     unsigned total_out = 0;
 
-    ui_pass = -1;  // Compressing Elf headers is invisible to UI.
-    x.offset = 0;
-    x.size = sizeof(Elf32_Ehdr) + sz_phdrs;
-    {
-        int const old_level = ph.level; ph.level = 10;
-        packExtent(x, total_in, total_out, 0, fo);
-        ph.level = old_level;
-    }
+    unsigned hdr_ulen = sizeof(Elf32_Ehdr) + sz_phdrs;
 
     ui_pass = 0;
     ft.addvalue = 0;
@@ -770,7 +763,8 @@ void PackLinuxElf32::pack2(OutputFile *fo, Filter &ft)
         // sometimes marks as PF_X anyway.  So filter only first segment.
         packExtent(x, total_in, total_out,
             ((0==nx && (Elf32_Phdr::PF_X & get_native32(&phdri[k].p_flags)))
-                ? &ft : 0 ), fo );
+                ? &ft : 0 ), fo, hdr_ulen);
+        hdr_ulen = 0;
         ++nx;
     }
     if (0!=ptload1sz && ptload0hi < ptload1lo) { // alignment hole?
@@ -819,14 +813,7 @@ void PackLinuxElf64::pack2(OutputFile *fo, Filter &ft)
     unsigned total_in = 0;
     unsigned total_out = 0;
 
-    ui_pass = -1;  // Compressing Elf headers is invisible to UI.
-    x.offset = 0;
-    x.size = sizeof(Elf64_Ehdr) + sz_phdrs;
-    {
-        int const old_level = ph.level; ph.level = 10;
-        packExtent(x, total_in, total_out, 0, fo);
-        ph.level = old_level;
-    }
+    unsigned hdr_ulen = sizeof(Elf64_Ehdr) + sz_phdrs;
 
     ui_pass = 0;
     ft.addvalue = 0;
@@ -851,7 +838,8 @@ void PackLinuxElf64::pack2(OutputFile *fo, Filter &ft)
         // sometimes marks as PF_X anyway.  So filter only first segment.
         packExtent(x, total_in, total_out,
             ((0==nx && (Elf64_Phdr::PF_X & get_native64(&phdri[k].p_flags)))
-                ? &ft : 0 ), fo );
+                ? &ft : 0 ), fo, hdr_ulen);
+        hdr_ulen = 0;
         ++nx;
     }
     if (0!=ptload1sz && ptload0hi < ptload1lo) { // alignment hole?
@@ -1513,14 +1501,7 @@ void PackLinuxI386elf::pack2(OutputFile *fo, Filter &ft)
     unsigned total_in = 0;
     unsigned total_out = 0;
 
-    ui_pass = -1;  // Compressing Elf headers is invisible to UI.
-    x.offset = 0;
-    x.size = sizeof(Elf32_Ehdr) + sz_phdrs;
-    {
-        int const old_level = ph.level; ph.level = 10;
-        packExtent(x, total_in, total_out, 0, fo);
-        ph.level = old_level;
-    }
+    unsigned hdr_ulen = sizeof(Elf32_Ehdr) + sz_phdrs;
 
     ui_pass = 0;
     ft.addvalue = 0;
@@ -1542,7 +1523,8 @@ void PackLinuxI386elf::pack2(OutputFile *fo, Filter &ft)
         }
         packExtent(x, total_in, total_out,
             ((Elf32_Phdr::PF_X & phdri[k].p_flags)
-                ? &ft : 0 ), fo );
+                ? &ft : 0 ), fo, hdr_ulen);
+        hdr_ulen = 0;
         ++nx;
     }
     if (0!=ptload1sz && ptload0hi < ptload1lo) { // alignment hole?
diff --git a/src/p_unix.cpp b/src/p_unix.cpp
index 57fcef93..ef41e949 100644
--- a/src/p_unix.cpp
+++ b/src/p_unix.cpp
@@ -304,9 +304,16 @@ void PackUnix::packExtent(
     unsigned &total_in,
     unsigned &total_out,
     Filter *ft,
-    OutputFile *fo
+    OutputFile *fo,
+    unsigned hdr_ulen
 )
 {
+    MemBuffer hdr_ibuf;
+    if (hdr_ulen) {
+        hdr_ibuf.alloc(hdr_ulen);
+        fi->seek(0, SEEK_SET);
+        int l = fi->readx(hdr_ibuf, hdr_ulen);
+    }
     fi->seek(x.offset, SEEK_SET);
     for (off_t rest = x.size; 0 != rest; ) {
         int const strategy = getStrategy(*ft);
@@ -336,7 +343,9 @@ void PackUnix::packExtent(
             ft->id = 0;
             ft->cto = 0;
 
-            compressWithFilters(ft, OVERHEAD, strategy);
+            compressWithFilters(ft, OVERHEAD, strategy,
+                NULL, 0, 0, 0, 0, // those 5 args are the defaults
+                hdr_ibuf, hdr_ulen);
         }
         else {
             (void) compress(ibuf, obuf);    // ignore return value
@@ -358,6 +367,24 @@ void PackUnix::packExtent(
 
         // write block sizes
         b_info tmp;
+        if (hdr_ulen) {
+            unsigned hdr_clen;
+            MemBuffer hdr_obuf;
+            unsigned result[16];
+            upx_compress_config_t conf;
+            memset(&conf, 0xff, sizeof(conf));
+            hdr_obuf.allocForCompression(hdr_ulen);
+            int r = upx_compress(hdr_ibuf, hdr_ulen, hdr_obuf, &hdr_clen, 0,
+                ph.method, 10, &conf, result);
+            memset(&tmp, 0, sizeof(tmp));
+            set_native32(&tmp.sz_unc, hdr_ulen);
+            set_native32(&tmp.sz_cpr, hdr_clen);
+            tmp.b_method = (unsigned char) ph.method;
+            fo->write(&tmp, sizeof(tmp));
+            b_len += sizeof(b_info);
+            fo->write(hdr_obuf, hdr_clen);
+            hdr_ulen = 0;  // compress hdr one time only
+        }
         memset(&tmp, 0, sizeof(tmp));
         set_native32(&tmp.sz_unc, ph.u_len);
         set_native32(&tmp.sz_cpr, ph.c_len);
diff --git a/src/p_unix.h b/src/p_unix.h
index 00d84232..b8227bec 100644
--- a/src/p_unix.h
+++ b/src/p_unix.h
@@ -81,7 +81,8 @@ protected:
         off_t size;
     };
     virtual void packExtent(const Extent &x,
-        unsigned &total_in, unsigned &total_out, Filter *, OutputFile *);
+        unsigned &total_in, unsigned &total_out, Filter *, OutputFile *,
+        unsigned hdr_len = 0);
     virtual void unpackExtent(unsigned wanted, OutputFile *fo,
         unsigned &total_in, unsigned &total_out,
         unsigned &c_adler, unsigned &u_adler,
diff --git a/src/packer.cpp b/src/packer.cpp
index dbf78cf9..2b459e90 100644
--- a/src/packer.cpp
+++ b/src/packer.cpp
@@ -1258,13 +1258,32 @@ const char *Packer::getDecompressor() const
 // executable formats.
 //
 // It will replace the tryFilters() / compress() call sequence.
+
+// 2006-02-15: hdr_buf and hdr_u_len are default empty input "header" array
+// to fix a 2-pass problem with Elf headers.  As of today there can be
+// only one decompression method per executable output file, and that method
+// is the one that gives best compression for .text and loader.  However,
+// the Elf headers precede .text in the output file, and are written first.
+// "--brute" compression often compressed the Elf headers using nrv2b
+// but the .text (and loader) with nrv2e.  This often resulted in SIGSEGV
+// during decompression.
+// The workaround is for hdr_buf and hdr_u_len to describe the Elf headers
+// (typically less than 512 bytes) when .text is passed in, and include
+// them in the calculation of shortest output.  Then the result
+// this->ph.method  will say which [single] method to use for everthing.
+// The Elf headers are never filtered.  They are short enough (< 512 bytes)
+// that compressing them more than once per method (once here when choosing,
+// once again just before writing [because compressWithFilters discards])
+// is OK because of the simplicity of not having two output arrays.
 **************************************************************************/
 
 void Packer::compressWithFilters(Filter *parm_ft,
                                  const unsigned overlap_range,
                                  int strategy, const int *parm_filters,
                                  unsigned max_offset, unsigned max_match,
-                                 unsigned filter_off, unsigned compress_buf_off)
+                                 unsigned filter_off, unsigned compress_buf_off,
+                                 unsigned char *hdr_buf,
+                                 unsigned hdr_u_len)
 {
     const int *f;
     //
@@ -1280,6 +1299,7 @@ void Packer::compressWithFilters(Filter *parm_ft,
     best_ph.c_len = orig_ph.u_len;
     best_ph.overlap_overhead = 0;
     unsigned best_ph_lsize = 0;
+    unsigned best_hdr_clen = 0;
 
     // preconditions
     assert(orig_ph.filter == 0);
@@ -1373,6 +1393,14 @@ void Packer::compressWithFilters(Filter *parm_ft,
     int nfilters_success = 0;
     for (int m = 0; m < nmethods; m++)          // for all methods
     {
+        unsigned hdr_clen = 0;
+        if (hdr_buf && hdr_u_len) {
+            unsigned result[16];
+            upx_compress_config_t conf;
+            memset(&conf, 0xff, sizeof(conf));
+            int r = upx_compress(hdr_buf, hdr_u_len, obuf, &hdr_clen,
+                0, methods[m], 10, &conf, result);
+        }
         for (int i = 0; i < nfilters; i++)          // for all filters
         {
             ibuf.checkState();
@@ -1428,19 +1456,19 @@ void Packer::compressWithFilters(Filter *parm_ft,
                     lsize = buildLoader(&ft);
                 }
 #if 0
-                printf("\n%2d %02x: %d +%4d = %d  (best: %d +%4d = %d)\n", ph.method, ph.filter,
-                       ph.c_len, lsize, ph.c_len + lsize,
-                       best_ph.c_len, best_ph_lsize, best_ph.c_len + best_ph_lsize);
+                printf("\n%2d %02x: %d +%4d +%3d = %d  (best: %d +%4d +%3d = %d)\n", ph.method, ph.filter,
+                       ph.c_len, lsize, hdr_clen, ph.c_len + lsize + hdr_clen,
+                       best_ph.c_len, best_ph_lsize, best_hdr_clen, best_ph.c_len + best_ph_lsize + best_hdr_clen);
 #endif
                 bool update = false;
-                if (ph.c_len + lsize < best_ph.c_len + best_ph_lsize)
+                if (ph.c_len + lsize + hdr_clen < best_ph.c_len + best_ph_lsize + best_hdr_clen)
                     update = true;
-                else if (ph.c_len + lsize == best_ph.c_len + best_ph_lsize)
+                else if (ph.c_len + lsize + hdr_clen == best_ph.c_len + best_ph_lsize + best_hdr_clen)
                 {
                     // prefer smaller loaders
-                    if (lsize < best_ph_lsize)
+                    if (lsize  + hdr_clen < best_ph_lsize + best_hdr_clen)
                         update = true;
-                    else if (lsize == best_ph_lsize)
+                    else if (lsize + hdr_clen == best_ph_lsize + best_hdr_clen)
                     {
                         // prefer less overlap_overhead
                         if (ph.overlap_overhead < best_ph.overlap_overhead)
@@ -1455,6 +1483,7 @@ void Packer::compressWithFilters(Filter *parm_ft,
                     // save compression results
                     best_ph = ph;
                     best_ph_lsize = lsize;
+                    best_hdr_clen = hdr_clen;
                     best_ft = ft;
                 }
             }
diff --git a/src/packer.h b/src/packer.h
index 4eda5e25..618f5f19 100644
--- a/src/packer.h
+++ b/src/packer.h
@@ -172,7 +172,9 @@ protected:
                              const int *filters = NULL,
                              unsigned max_offset = 0, unsigned max_match = 0,
                              unsigned filter_buf_off = 0,
-                             unsigned compress_buf_off = 0);
+                             unsigned compress_buf_off = 0,
+                             unsigned char *header_buffer = 0,
+                             unsigned header_length = 0);
 
     // util for verifying overlapping decompresion
     //   non-destructive test