diff --git a/fft.py b/fft.py
index f533312..acfc3e8 100644
--- a/fft.py
+++ b/fft.py
@@ -77,10 +77,11 @@ def real_fft( f,inv ):
         
         F1k = Fp[k] + Fp[N-k].conjugate()
         F2k = Fp[k] - Fp[N-k].conjugate()
-        F[k] = ( F1k + tw * F2k ) * .5
-        
+        F2k *= tw
+        F[k] = ( F1k + F2k ) * .5
+        F[N-k] = ( F1k - F2k ).conjugate() * .5
         #F[N-k] = ( F1kp + e ** ( -j*pi*(.5+float(N-k)/N ) ) * F2kp ) * .5
-        F[N-k] = ( F1k.conjugate() - tw.conjugate() * F2k.conjugate() ) * .5
+        #F[N-k] = ( F1k.conjugate() - tw.conjugate() * F2k.conjugate() ) * .5
 
     F[N] = complex( Fp[0].real - Fp[0].imag , 0 ) 
     return F
diff --git a/kiss_fftr.c b/kiss_fftr.c
index c76cca7..b52539a 100644
--- a/kiss_fftr.c
+++ b/kiss_fftr.c
@@ -79,25 +79,30 @@ void kiss_fftr(const void * cfg,const kiss_fft_scalar *timedata,kiss_fft_cpx *fr
     freqdata[0].r = st->tmpbuf[0].r + st->tmpbuf[0].i;
     freqdata[0].i = 0;
 
-    for (k=1;k<N;++k) {
-        kiss_fft_cpx fpnk,fpk,f1k;
+    for (k=1;k <= N/2 ; ++k ) {
+        kiss_fft_cpx fpnk,fpk,f1k,f2k,tw;
 
         fpk = st->tmpbuf[k]; 
         fpnk.r =  st->tmpbuf[N-k].r;
         fpnk.i = -st->tmpbuf[N-k].i;
 
         C_ADD( f1k, fpk , fpnk );
-        C_SUBFROM( fpk , fpnk );
-        C_MUL( freqdata[k], fpk , st->super_twiddles[k] );
-        C_ADDTO(freqdata[k],f1k);
+        C_SUB( f2k, fpk , fpnk );
+        C_MUL( tw , f2k , st->super_twiddles[k]);
 
+        C_ADD( freqdata[k] , f1k ,tw);
         freqdata[k].r /= 2;
         freqdata[k].i /= 2;
+
+        freqdata[N-k].r =    f1k.r - tw.r;
+        freqdata[N-k].i = - (f1k.i - tw.i);
+        freqdata[N-k].r /= 2;
+        freqdata[N-k].i /= 2;
     }
     freqdata[N].r = st->tmpbuf[0].r - st->tmpbuf[0].i;
     freqdata[N].i = 0;
 }
-    
+
 void kiss_fftri(const void * cfg,const kiss_fft_cpx *freqdata,kiss_fft_scalar *timedata)
 {
     /* input buffer timedata is stored row-wise */
diff --git a/test/test_real.c b/test/test_real.c
index 5ccf4d2..cf17a5e 100644
--- a/test/test_real.c
+++ b/test/test_real.c
@@ -34,9 +34,7 @@ double snr_compare( kiss_fft_cpx * vec1,kiss_fft_cpx * vec2, int n)
         printf( "\npoor snr, try a scaling factor %f\n" , scale );
     return snr;
 }
-
 #define RANDOM
- 
 #ifndef RANDOM
 #define NFFT 8
 #else
@@ -80,6 +78,22 @@ int main()
     kiss_fftr(kiss_fftr_state,sin,sout);
     printf( "nfft=%d, inverse=%d, snr=%g\n",
             NFFT,0, snr_compare(cout,sout,(NFFT/2)+1) );
+#ifdef RANDOM        
+    ts = cputime();
+    for (i=0;i<NUMFFTS;++i) {
+        kiss_fft(kiss_fft_state,cin,cout);
+    }
+    tfft = cputime() - ts;
+    
+    ts = cputime();
+    for (i=0;i<NUMFFTS;++i) {
+        kiss_fftr( kiss_fftr_state, sin, cout );
+        /* kiss_fftri(kiss_fftr_state,cout,sin); */
+    }
+    trfft = cputime() - ts;
+
+    printf("%d complex ffts took %gs, real took %gs\n",NUMFFTS,tfft,trfft);
+#endif        
     free(kiss_fft_state);
     free(kiss_fftr_state);
 
@@ -99,22 +113,6 @@ int main()
     
     printf( "nfft=%d, inverse=%d, snr=%g\n",
             NFFT,1, snr_compare(cin,cin,NFFT/2) );
-#ifdef RANDOM        
-    ts = cputime();
-    for (i=0;i<NUMFFTS;++i) {
-        kiss_fft(kiss_fft_state,cin,cout);
-    }
-    tfft = cputime() - ts;
-    
-    ts = cputime();
-    for (i=0;i<NUMFFTS;++i) {
-        /* kiss_fftr(kiss_fftr_state,sin,cout); */
-        kiss_fftri(kiss_fftr_state,cout,sin);
-    }
-    trfft = cputime() - ts;
-
-    printf("%d complex ffts took %gs, real took %gs\n",NUMFFTS,tfft,trfft);
-#endif        
     free(kiss_fft_state);
     free(kiss_fftr_state);