diff --git a/libraries/SD/src/SD.cpp b/libraries/SD/src/SD.cpp
index 65d32741c..d426dbe8e 100644
--- a/libraries/SD/src/SD.cpp
+++ b/libraries/SD/src/SD.cpp
@@ -326,13 +326,7 @@ boolean callback_rmdir(SdFile& parentDir, char *filePathComponent,
   return true;
 }
 
-
-
-/* Implementation of class used to create `SDCard` object. */
-
-
-
-boolean SDClass::begin(uint8_t csPin) {
+boolean SDClass::begin(uint8_t csPin, uint32_t sckRateID) {
   /*
 
     Performs the initialisation required by the sdfatlib library.
@@ -340,13 +334,11 @@ boolean SDClass::begin(uint8_t csPin) {
     Return true if initialization succeeds, false otherwise.
 
    */
-  return card.init(SPI_HALF_SPEED, csPin) &&
+  return card.init(sckRateID, csPin) &&
          volume.init(card) &&
          root.openRoot(volume);
 }
 
-
-
 // this little helper is used to traverse paths
 SdFile SDClass::getParentDir(const char *filepath, int *index) {
   // get parent directory
diff --git a/libraries/SD/src/SD.h b/libraries/SD/src/SD.h
index 7435cf577..e229cf8ad 100644
--- a/libraries/SD/src/SD.h
+++ b/libraries/SD/src/SD.h
@@ -65,8 +65,8 @@ private:
 public:
   // This needs to be called to set up the connection to the SD card
   // before other methods are used.
-  boolean begin(uint8_t csPin = SD_CHIP_SELECT_PIN);
-  
+  boolean begin(uint8_t csPin = SD_CHIP_SELECT_PIN, uint32_t sckRateID = SPI_FULL_SPEED);
+
   // Open the specified file/directory with the supplied mode (e.g. read or
   // write, etc). Returns a File object for interacting with the file.
   // Note that currently only one file can be open at a time.
diff --git a/libraries/SD/src/utility/Sd2Card.cpp b/libraries/SD/src/utility/Sd2Card.cpp
index 98a02ba42..2d7618d65 100644
--- a/libraries/SD/src/utility/Sd2Card.cpp
+++ b/libraries/SD/src/utility/Sd2Card.cpp
@@ -33,9 +33,13 @@ static void spiSend(uint8_t b) {
   SPDR = b;
   while (!(SPSR & (1 << SPIF)))
     ;
+#else
+#ifdef ESP8266
+  SPI.write(b);
 #else
   SPI.transfer(b);
 #endif
+#endif
 }
 /** Receive a byte from the card */
 static  uint8_t spiRec(void) {
@@ -116,8 +120,14 @@ uint8_t Sd2Card::cardCommand(uint8_t cmd, uint32_t arg) {
   // send command
   spiSend(cmd | 0x40);
 
+#ifdef ESP8266
+  // send argument
+  SPI.write32(arg, true);
+#else
   // send argument
   for (int8_t s = 24; s >= 0; s -= 8) spiSend(arg >> s);
+#endif
+
 
   // send CRC
   uint8_t crc = 0xFF;
@@ -424,7 +434,14 @@ uint8_t Sd2Card::readData(uint32_t block,
   dst[n] = SPDR;
 
 #else  // OPTIMIZE_HARDWARE_SPI
+#ifdef ESP8266
+  // skip data before offset
+  SPI.transferBytes(NULL, NULL, offset_);
 
+  // transfer data
+  SPI.transferBytes(NULL, dst, count);
+
+#else
   // skip data before offset
   for (;offset_ < offset; offset_++) {
     spiRec();
@@ -433,6 +450,7 @@ uint8_t Sd2Card::readData(uint32_t block,
   for (uint16_t i = 0; i < count; i++) {
     dst[i] = spiRec();
   }
+#endif
 #endif  // OPTIMIZE_HARDWARE_SPI
 
   offset_ += count;
@@ -463,7 +481,11 @@ void Sd2Card::readEnd(void) {
     while (!(SPSR & (1 << SPIF)))
       ;
 #else  // OPTIMIZE_HARDWARE_SPI
+#ifdef ESP8266
+    SPI.transferBytes(NULL, NULL, (514-offset_));
+#else
     while (offset_++ < 514) spiRec();
+#endif
 #endif  // OPTIMIZE_HARDWARE_SPI
     chipSelectHigh();
     inBlock_ = 0;
@@ -479,7 +501,11 @@ uint8_t Sd2Card::readRegister(uint8_t cmd, void* buf) {
   }
   if (!waitStartBlock()) goto fail;
   // transfer data
+#ifdef ESP8266
+  SPI.transferBytes(NULL, dst, 16);
+#else
   for (uint16_t i = 0; i < 16; i++) dst[i] = spiRec();
+#endif
   spiRec();  // get first crc byte
   spiRec();  // get second crc byte
   chipSelectHigh();
@@ -646,13 +672,21 @@ uint8_t Sd2Card::writeData(uint8_t token, const uint8_t* src) {
 
 #else  // OPTIMIZE_HARDWARE_SPI
   spiSend(token);
+#ifdef ESP8266
+  // send argument
+  SPI.writeBytes((uint8_t *)src, 512);
+#else
   for (uint16_t i = 0; i < 512; i++) {
     spiSend(src[i]);
   }
+#endif
 #endif  // OPTIMIZE_HARDWARE_SPI
+#ifdef ESP8266
+  SPI.write16(0xFFFF, true);
+#else
   spiSend(0xff);  // dummy crc
   spiSend(0xff);  // dummy crc
-
+#endif
   status_ = spiRec();
   if ((status_ & DATA_RES_MASK) != DATA_RES_ACCEPTED) {
     error(SD_CARD_ERROR_WRITE);
diff --git a/libraries/SPI/SPI.cpp b/libraries/SPI/SPI.cpp
index 6b3d29784..be4627740 100644
--- a/libraries/SPI/SPI.cpp
+++ b/libraries/SPI/SPI.cpp
@@ -216,8 +216,16 @@ void SPIClass::setClockDivider(uint32_t clockDiv) {
     SPI1CLK = clockDiv;
 }
 
+inline void SPIClass::setDataBits(uint16_t bits) {
+    const uint32_t mask = ~((SPIMMOSI << SPILMOSI) | (SPIMMISO << SPILMISO));
+    bits--;
+    SPI1U1 = ((SPI1U1 & mask) | ((bits << SPILMOSI) | (bits << SPILMISO)));
+}
+
 uint8_t SPIClass::transfer(uint8_t data) {
     while(SPI1CMD & SPIBUSY) {}
+    // reset to 8Bit mode
+    setDataBits(8);
     SPI1W0 = data;
     SPI1CMD |= SPIBUSY;
     while(SPI1CMD & SPIBUSY) {}
@@ -248,19 +256,21 @@ uint16_t SPIClass::transfer16(uint16_t data) {
 
 void SPIClass::write(uint8_t data) {
     while(SPI1CMD & SPIBUSY) {}
+    // reset to 8Bit mode
+    setDataBits(8);
     SPI1W0 = data;
     SPI1CMD |= SPIBUSY;
     while(SPI1CMD & SPIBUSY) {}
 }
 
 void SPIClass::write16(uint16_t data) {
-    write16(data, (SPI1C & (SPICWBO | SPICRBO)));
+    write16(data, !(SPI1C & (SPICWBO | SPICRBO)));
 }
 
 void SPIClass::write16(uint16_t data, bool msb) {
     while(SPI1CMD & SPIBUSY) {}
     // Set to 16Bits transfer
-    SPI1U1 = (SPI1U1 & ~((SPIMMOSI << SPILMOSI))) | ((16 - 1) << SPILMOSI);
+    setDataBits(16);
     if(msb) {
         // MSBFIRST Byte first
         SPI1W0 = (data >> 8) | (data << 8);
@@ -271,8 +281,31 @@ void SPIClass::write16(uint16_t data, bool msb) {
         SPI1CMD |= SPIBUSY;
     }
     while(SPI1CMD & SPIBUSY) {}
-    // reset to 8Bit mode
-    SPI1U1 = (SPI1U1 & ~((SPIMMOSI << SPILMOSI))) | ((8 - 1) << SPILMOSI);
+}
+
+void SPIClass::write32(uint32_t data) {
+    write32(data, !(SPI1C & (SPICWBO | SPICRBO)));
+}
+
+void SPIClass::write32(uint32_t data, bool msb) {
+    while(SPI1CMD & SPIBUSY) {}
+    // Set to 32Bits transfer
+    setDataBits(32);
+    if(msb) {
+        union {
+                uint32_t l;
+                uint8_t b[4];
+        } data_;
+        data_.l = data;
+        // MSBFIRST Byte first
+        SPI1W0 = (data_.b[3] | (data_.b[2] << 8) | (data_.b[1] << 16) | (data_.b[0] << 24));
+        SPI1CMD |= SPIBUSY;
+    } else {
+        // LSBFIRST Byte first
+        SPI1W0 = data;
+        SPI1CMD |= SPIBUSY;
+    }
+    while(SPI1CMD & SPIBUSY) {}
 }
 
 void SPIClass::writeBytes(uint8_t * data, uint32_t size) {
@@ -291,7 +324,7 @@ void SPIClass::writeBytes(uint8_t * data, uint32_t size) {
 void SPIClass::writeBytes_(uint8_t * data, uint8_t size) {
     while(SPI1CMD & SPIBUSY) {}
     // Set Bits to transfer
-    SPI1U1 = (SPI1U1 & ~((SPIMMOSI << SPILMOSI))) | ((size * 8 - 1) << SPILMOSI);
+    setDataBits(size * 8);
 
     volatile uint32_t * fifoPtr = &SPI1W0;
     uint32_t * dataPtr = (uint32_t*) data;
@@ -305,8 +338,6 @@ void SPIClass::writeBytes_(uint8_t * data, uint8_t size) {
 
     SPI1CMD |= SPIBUSY;
     while(SPI1CMD & SPIBUSY) {}
-    // reset to 8Bit mode
-    SPI1U1 = (SPI1U1 & ~((SPIMMOSI << SPILMOSI))) | ((8 - 1) << SPILMOSI);
 }
 
 void SPIClass::writePattern(uint8_t * data, uint8_t size, uint32_t repeat) {
@@ -344,3 +375,56 @@ void SPIClass::writePattern_(uint8_t * data, uint8_t size, uint8_t repeat) {
 
     writeBytes(&buffer[0], bytes);
 }
+
+void SPIClass::transferBytes(uint8_t * out, uint8_t * in, uint32_t size) {
+    while(size) {
+        if(size > 64) {
+            transferBytes_(out, in, 64);
+            size -= 64;
+            if(out) out += 64;
+            if(in) in += 64;
+        } else {
+            transferBytes_(out, in, size);
+            size = 0;
+        }
+    }
+}
+
+void SPIClass::transferBytes_(uint8_t * out, uint8_t * in, uint8_t size) {
+    while(SPI1CMD & SPIBUSY) {}
+    // Set in/out Bits to transfer
+
+    setDataBits(size * 8);
+
+    volatile uint32_t * fifoPtr = &SPI1W0;
+    uint8_t dataSize = ((size + 3) / 4);
+
+    if(out) {
+        uint32_t * dataPtr = (uint32_t*) out;
+        while(dataSize--) {
+            *fifoPtr = *dataPtr;
+            dataPtr++;
+            fifoPtr++;
+        }
+    } else {
+        // no out data only read fill with dummy data!
+        while(dataSize--) {
+            *fifoPtr = 0xFFFFFFFF;
+            fifoPtr++;
+        }
+    }
+
+    SPI1CMD |= SPIBUSY;
+    while(SPI1CMD & SPIBUSY) {}
+
+    if(in) {
+        volatile uint8_t * fifoPtr8 = (volatile uint8_t *) &SPI1W0;
+        dataSize = size;
+        while(dataSize--) {
+            *in = *fifoPtr8;
+            in++;
+            fifoPtr8++;
+        }
+    }
+}
+
diff --git a/libraries/SPI/SPI.h b/libraries/SPI/SPI.h
index 1030eeaa9..e67b5b0d5 100644
--- a/libraries/SPI/SPI.h
+++ b/libraries/SPI/SPI.h
@@ -75,13 +75,18 @@ public:
   void write(uint8_t data);
   void write16(uint16_t data);
   void write16(uint16_t data, bool msb);
+  void write32(uint32_t data);
+  void write32(uint32_t data, bool msb);
   void writeBytes(uint8_t * data, uint32_t size);
   void writePattern(uint8_t * data, uint8_t size, uint32_t repeat);
+  void transferBytes(uint8_t * out, uint8_t * in, uint32_t size);
   void endTransaction(void);
 private:
   bool useHwCs;
   void writeBytes_(uint8_t * data, uint8_t size);
   void writePattern_(uint8_t * data, uint8_t size, uint8_t repeat);
+  void transferBytes_(uint8_t * out, uint8_t * in, uint8_t size);
+  inline void setDataBits(uint16_t bits);
 };
 
 extern SPIClass SPI;