diff --git a/micropython/drivers/storage/sdcard/manifest.py b/micropython/drivers/storage/sdcard/manifest.py
index 2589343d4..37fae6218 100644
--- a/micropython/drivers/storage/sdcard/manifest.py
+++ b/micropython/drivers/storage/sdcard/manifest.py
@@ -1,3 +1,3 @@
-metadata(description="SDCard block device driver.", version="0.2.0")
+metadata(description="SDCard block device driver.", version="0.2.1")
 
 module("sdcard.py", opt=3)
diff --git a/micropython/drivers/storage/sdcard/sdcard.py b/micropython/drivers/storage/sdcard/sdcard.py
index 992f52eea..8f4372586 100644
--- a/micropython/drivers/storage/sdcard/sdcard.py
+++ b/micropython/drivers/storage/sdcard/sdcard.py
@@ -104,7 +104,7 @@ def init_card(self, baudrate):
         if self.cmd(9, 0, 0, False) != 0:
             raise OSError("no response from SD card")
         csd = bytearray(16)
-        self.readinto(csd)
+        self.readinto(csd, True)
         if csd[0] & 0xC0 == 0x40:  # CSD version 2.0
             self.sectors = ((csd[7] << 16 | csd[8] << 8 | csd[9]) + 1) * 1024
         elif csd[0] & 0xC0 == 0x00:  # CSD version 1.0 (old, <=2GB)
@@ -191,7 +191,7 @@ def cmd(self, cmd, arg, final=0, release=True, skip1=False):
         self.spi.write(b"\xff")
         return -1
 
-    def readinto(self, buf):
+    def readinto(self, buf, release):
         self.cs(0)
 
         # read until start byte (0xff)
@@ -214,10 +214,11 @@ def readinto(self, buf):
         self.spi.write(b"\xff")
         self.spi.write(b"\xff")
 
-        self.cs(1)
-        self.spi.write(b"\xff")
+        if release:
+            self.cs(1)
+            self.spi.write(b"\xff")
 
-    def write(self, token, buf):
+    def write(self, token, buf, release):
         self.cs(0)
 
         # send: start of block, data, checksum
@@ -236,8 +237,9 @@ def write(self, token, buf):
         while self.spi.read(1, 0xFF)[0] == 0:
             pass
 
-        self.cs(1)
-        self.spi.write(b"\xff")
+        if release:
+            self.cs(1)
+            self.spi.write(b"\xff")
 
     def write_token(self, token):
         self.cs(0)
@@ -264,7 +266,7 @@ def readblocks(self, block_num, buf):
                 self.cs(1)
                 raise OSError(5)  # EIO
             # receive the data and release card
-            self.readinto(buf)
+            self.readinto(buf, True)
         else:
             # CMD18: set read address for multiple blocks
             if self.cmd(18, block_num * self.cdv, release=False) != 0:
@@ -274,8 +276,10 @@ def readblocks(self, block_num, buf):
             offset = 0
             mv = memoryview(buf)
             while nblocks:
-                # receive the data and release card
-                self.readinto(mv[offset : offset + 512])
+                # receive the data, keeping the card selected until CMD12:
+                # some cards lose bit alignment if CS is released between
+                # the blocks of a multi-block read
+                self.readinto(mv[offset : offset + 512], False)
                 offset += 512
                 nblocks -= 1
             if self.cmd(12, 0, skip1=True):
@@ -294,16 +298,18 @@ def writeblocks(self, block_num, buf):
                 raise OSError(5)  # EIO
 
             # send the data
-            self.write(_TOKEN_DATA, buf)
+            self.write(_TOKEN_DATA, buf, True)
         else:
             # CMD25: set write address for first block
-            if self.cmd(25, block_num * self.cdv) != 0:
+            if self.cmd(25, block_num * self.cdv, release=False) != 0:
+                # release the card
+                self.cs(1)
                 raise OSError(5)  # EIO
-            # send the data
+            # send the data, keeping the card selected until the stop token
             offset = 0
             mv = memoryview(buf)
             while nblocks:
-                self.write(_TOKEN_CMD25, mv[offset : offset + 512])
+                self.write(_TOKEN_CMD25, mv[offset : offset + 512], False)
                 offset += 512
                 nblocks -= 1
             self.write_token(_TOKEN_STOP_TRAN)