diff --git a/scripts/lsriff.py b/scripts/lsriff.py
index 2c08e3c..2c0640e 100755
--- a/scripts/lsriff.py
+++ b/scripts/lsriff.py
@@ -2,7 +2,7 @@
 
 import sys
 
-from yoga.image.helpers import get_riff_structure
+from yoga.image.encoders.webp import get_riff_structure
 
 
 def print_riff_info(input_path):
diff --git a/test/test_image_encoders_webp.py b/test/test_image_encoders_webp.py
new file mode 100644
index 0000000..97407cd
--- /dev/null
+++ b/test/test_image_encoders_webp.py
@@ -0,0 +1,28 @@
+import pytest
+
+from yoga.image.encoders import webp
+
+
+class Test_little_endian_unint32_bytes_to_python_int(object):
+    def test_uint32_value(self):
+        assert (
+            webp.little_endian_unint32_bytes_to_python_int(b"\x78\x56\x34\x12")
+            == 305419896
+        )
+
+
+class Test_get_riff_structure(object):
+    @pytest.fixture
+    def webp_image(self):
+        return open("test/images/alpha.lossless.metadata.webp", "rb").read()
+
+    def test_riff_structure(object, webp_image):
+        riff = webp.get_riff_structure(webp_image)
+        assert riff["formtype"] == "WEBP"
+        assert riff["size"] == 11868
+        assert len(riff["chunks"]) == 5
+        assert riff["chunks"][0]["type"] == "VP8X"
+        assert riff["chunks"][1]["type"] == "ICCP"
+        assert riff["chunks"][2]["type"] == "VP8L"
+        assert riff["chunks"][3]["type"] == "EXIF"
+        assert riff["chunks"][4]["type"] == "XMP "
diff --git a/test/test_image_helpers.py b/test/test_image_helpers.py
index 7231ba0..290eec8 100644
--- a/test/test_image_helpers.py
+++ b/test/test_image_helpers.py
@@ -37,33 +37,6 @@ class Test_image_have_alpha(object):
             assert not helpers.image_have_alpha(image, threshold)
 
 
-class Test_little_endian_unint32_bytes_to_python_int(object):
-    def test_uint32_value(self):
-        assert (
-            helpers.little_endian_unint32_bytes_to_python_int(
-                b"\x78\x56\x34\x12"
-            )
-            == 305419896
-        )
-
-
-class Test_get_riff_structure(object):
-    @pytest.fixture
-    def webp_image(self):
-        return open("test/images/alpha.lossless.metadata.webp", "rb").read()
-
-    def test_riff_structure(object, webp_image):
-        riff = helpers.get_riff_structure(webp_image)
-        assert riff["formtype"] == "WEBP"
-        assert riff["size"] == 11868
-        assert len(riff["chunks"]) == 5
-        assert riff["chunks"][0]["type"] == "VP8X"
-        assert riff["chunks"][1]["type"] == "ICCP"
-        assert riff["chunks"][2]["type"] == "VP8L"
-        assert riff["chunks"][3]["type"] == "EXIF"
-        assert riff["chunks"][4]["type"] == "XMP "
-
-
 class Test_guess_image_format(object):
     @pytest.mark.parametrize(
         "image_path, expected_format",
diff --git a/yoga/image/__init__.py b/yoga/image/__init__.py
index 0474c5d..7b51755 100644
--- a/yoga/image/__init__.py
+++ b/yoga/image/__init__.py
@@ -125,7 +125,8 @@ API
 
 from PIL import Image
 
-from . import encoders
+from .encoders.jpeg import optimize_jpeg
+from .encoders.png import optimize_png
 from .options import normalize_options
 from .helpers import image_have_alpha
 
@@ -173,9 +174,9 @@ def optimize(input_file, output_file, options={}, verbose=False, quiet=False):
     # convert / optimize
     output_image_bytes = None
     if output_format == "jpeg":
-        output_image_bytes = encoders.jpeg(image, options["jpeg_quality"])
+        output_image_bytes = optimize_jpeg(image, options["jpeg_quality"])
     elif output_format == "png":
-        output_image_bytes = encoders.png(image)
+        output_image_bytes = optimize_png(image)
     else:
         raise ValueError("Invalid output format %s" % output_format)
 
diff --git a/yoga/image/encoders/__init__.py b/yoga/image/encoders/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/yoga/image/encoders/jpeg.py b/yoga/image/encoders/jpeg.py
new file mode 100644
index 0000000..d0146fa
--- /dev/null
+++ b/yoga/image/encoders/jpeg.py
@@ -0,0 +1,34 @@
+import pyguetzli
+
+
+def is_jpeg(file_bytes):
+    """Whether or not the given bytes represent a JPEG file.
+
+    :params bytes file_bytes: The bytes of the file to check.
+
+    :rtype: bool
+    :return: ``True`` if the bytes represent a JPEG file, ``False`` else.
+    """
+    JPEG_MAGICS = [
+        b"\xFF\xD8\xFF\xDB",
+        b"\xFF\xD8\xFF\xE0\x00\x10\x4A\x46\x49\x46\x00\x01",  # JFIF format
+        b"\xFF\xD8\xFF\xEE",
+        b"\xFF\xD8\xFF\xE1",  # xx xx 45 78 69 66 00 00  / Exif format
+    ]
+    for magic in JPEG_MAGICS:
+        if file_bytes.startswith(magic):
+            return True
+    return False
+
+
+def optimize_jpeg(image, quality):
+    """Encode image to JPEG using Guetzli.
+
+    :param PIL.Image image: The image to encode.
+    :param float quality: The output JPEG quality (from ``0.00``to ``1.00``).
+
+    :returns: The encoded image's bytes.
+    """
+    if not 0.00 <= quality <= 1.00:
+        raise ValueError("JPEG quality value must be between 0.00 and 1.00")
+    return pyguetzli.process_pil_image(image, int(quality * 100))
diff --git a/yoga/image/encoders.py b/yoga/image/encoders/png.py
similarity index 59%
rename from yoga/image/encoders.py
rename to yoga/image/encoders/png.py
index efa70ae..cb8079e 100644
--- a/yoga/image/encoders.py
+++ b/yoga/image/encoders/png.py
@@ -1,23 +1,20 @@
 import io
 
-import pyguetzli
 import zopfli
 
 
-def jpeg(image, quality):
-    """Encode image to JPEG using Guetzli.
+def is_png(file_bytes):
+    """Whether or not the given bytes represent a PNG file.
 
-    :param PIL.Image image: The image to encode.
-    :param float quality: The output JPEG quality (from ``0.00``to ``1.00``).
+    :params bytes file_bytes: The bytes of the file to check.
 
-    :returns: The encoded image's bytes.
+    :rtype: bool
+    :return: ``True`` if the bytes represent a PNG file, ``False`` else.
     """
-    if not 0.00 <= quality <= 1.00:
-        raise ValueError("JPEG quality value must be between 0.00 and 1.00")
-    return pyguetzli.process_pil_image(image, int(quality * 100))
+    return file_bytes.startswith(b"\x89PNG\r\n")
 
 
-def png(image):
+def optimize_png(image):
     """Encode image to PNG using ZopfliPNG.
 
     :param PIL.Image image: The image to encode.
diff --git a/yoga/image/encoders/webp.py b/yoga/image/encoders/webp.py
new file mode 100644
index 0000000..b765b64
--- /dev/null
+++ b/yoga/image/encoders/webp.py
@@ -0,0 +1,66 @@
+import struct
+
+
+def little_endian_unint32_bytes_to_python_int(bytes_):
+    return struct.unpack("<L", bytes_)[0]
+
+
+def get_riff_structure(data):
+    if data[0:4] != b"RIFF":
+        raise ValueError("Unvalid RIFF: Not a RIFF file")
+
+    result = {
+        "formtype": data[8:12].decode(),
+        "size": little_endian_unint32_bytes_to_python_int(data[4:8]),
+        "chunks": [],
+    }
+
+    if result["size"] + 8 != len(data):
+        raise ValueError("Unvalid RIFF: Truncated data")
+
+    offset = 12  # RIFF header length
+
+    while offset < len(data):
+        chunk = {
+            "type": data[offset : offset + 4].decode(),
+            "data_offset": offset + 8,
+            "size": little_endian_unint32_bytes_to_python_int(
+                data[offset + 4 : offset + 8]
+            ),
+        }
+        result["chunks"].append(chunk)
+        offset += 8 + chunk["size"] + chunk["size"] % 2
+
+    return result
+
+
+def is_riff(file_bytes):
+    """Whether or not the given bytes represent a RIFF file.
+
+    :params bytes file_bytes: The bytes of the file to check.
+
+    :rtype: bool
+    :return: ``True`` if the bytes represent a RIFF file, ``False`` else.
+    """
+    return file_bytes.startswith(b"RIFF")
+
+
+def is_lossy_webp(file_bytes):
+    """Whether or not the given bytes represent a lossy WEBP file.
+
+    :params bytes file_bytes: The bytes of the file to check.
+
+    :rtype: bool
+    :return: ``True`` if the bytes represent a lossy WEBP file, ``False`` else.
+    """
+    if not is_riff(file_bytes):
+        return False
+
+    riff = get_riff_structure(file_bytes)
+
+    if riff["formtype"] == "WEBP":
+        chunks = [chunk["type"] for chunk in riff["chunks"]]
+        if "VP8 " in chunks:
+            return True
+
+    return False
diff --git a/yoga/image/encoders/webp_lossless.py b/yoga/image/encoders/webp_lossless.py
new file mode 100644
index 0000000..e5069f5
--- /dev/null
+++ b/yoga/image/encoders/webp_lossless.py
@@ -0,0 +1,23 @@
+from .webp import is_riff, get_riff_structure
+
+
+def is_lossless_webp(file_bytes):
+    """Whether or not the given bytes represent a lossless WEBP file.
+
+    :params bytes file_bytes: The bytes of the file to check.
+
+    :rtype: bool
+    :return: ``True`` if the bytes represent a lossless WEBP file, ``False``
+             else.
+    """
+    if not is_riff(file_bytes):
+        return False
+
+    riff = get_riff_structure(file_bytes)
+
+    if riff["formtype"] == "WEBP":
+        chunks = [chunk["type"] for chunk in riff["chunks"]]
+        if "VP8L" in chunks:
+            return True
+
+    return False
diff --git a/yoga/image/helpers.py b/yoga/image/helpers.py
index 4798de8..612a219 100644
--- a/yoga/image/helpers.py
+++ b/yoga/image/helpers.py
@@ -1,4 +1,7 @@
-import struct
+from .encoders.jpeg import is_jpeg
+from .encoders.png import is_png
+from .encoders.webp import is_lossy_webp
+from .encoders.webp_lossless import is_lossless_webp
 
 
 def image_have_alpha(image, threshold=0xFE):
@@ -14,53 +17,16 @@ def image_have_alpha(image, threshold=0xFE):
     return False
 
 
-def little_endian_unint32_bytes_to_python_int(bytes_):
-    return struct.unpack("<L", bytes_)[0]
-
-
-def get_riff_structure(data):
-    if data[0:4] != b"RIFF":
-        raise ValueError("Unvalid RIFF: Not a RIFF file")
-
-    result = {
-        "formtype": data[8:12].decode(),
-        "size": little_endian_unint32_bytes_to_python_int(data[4:8]),
-        "chunks": [],
+def guess_image_format(image_bytes):
+    FORMATS = {
+        "jpeg": is_jpeg,
+        "png": is_png,
+        "webp": is_lossy_webp,
+        "webpl": is_lossless_webp,
     }
 
-    if result["size"] + 8 != len(data):
-        raise ValueError("Unvalid RIFF: Truncated data")
-
-    offset = 12  # RIFF header length
-
-    while offset < len(data):
-        chunk = {
-            "type": data[offset : offset + 4].decode(),
-            "data_offset": offset + 8,
-            "size": little_endian_unint32_bytes_to_python_int(
-                data[offset + 4 : offset + 8]
-            ),
-        }
-        result["chunks"].append(chunk)
-        offset += 8 + chunk["size"] + chunk["size"] % 2
-
-    return result
-
-
-def guess_image_format(image_bytes):
-    if image_bytes.startswith(b"\xFF\xD8\xFF\xE0"):
-        return "jpeg"
-
-    if image_bytes.startswith(b"\x89PNG\r\n"):
-        return "png"
-
-    if image_bytes.startswith(b"RIFF"):
-        riff = get_riff_structure(image_bytes)
-        if riff["formtype"] == "WEBP":
-            chunks = [chunk["type"] for chunk in riff["chunks"]]
-            if "VP8 " in chunks:
-                return "webp"
-            if "VP8L" in chunks:
-                return "webpl"
+    for format_, checker in FORMATS.items():
+        if checker(image_bytes):
+            return format_
 
     raise ValueError("Unsupported image format")