diff --git a/swh/loader/npm/client.py b/swh/loader/npm/client.py
--- a/swh/loader/npm/client.py
+++ b/swh/loader/npm/client.py
@@ -3,17 +3,15 @@
 # License: GNU General Public License version 3, or any later version
 # See top-level LICENSE file for more information
 
-import json
 import logging
 import os
 
-import chardet
 import requests
 
 from swh.core import tarball
 from swh.model import hashutil
 
-from swh.loader.npm.utils import extract_npm_package_author
+from swh.loader.npm.utils import extract_npm_package_author, load_json
 
 
 class NpmClient:
@@ -185,14 +183,12 @@
 
         # do not archive useless tarball root directory
         package_path = os.path.join(path, 'package')
-        # some old packages use their name as root directory
+        # some old packages use a root directory with a different name
         if not os.path.exists(package_path):
-            ver_pos = package_source_data['filename'].rfind(version)
-            package_name = package_source_data['filename'][:ver_pos-1]
-            package_path = os.path.join(path, package_name)
-        # fallback: archive root tarball directory
-        if not os.path.exists(package_path):
-            package_path = path
+            for _, dirnames, _ in os.walk(path):
+                if dirnames:
+                    package_path = os.path.join(path, dirnames[0])
+                break
 
         self.log.debug('Package local path: %s', package_path)
 
@@ -203,8 +199,7 @@
         package_json = {}
         with open(package_json_path, 'rb') as package_json_file:
             package_json_bytes = package_json_file.read()
-            file_encoding = chardet.detect(package_json_bytes)['encoding']
-            package_json = json.loads(package_json_bytes.decode(file_encoding))
+            package_json = load_json(package_json_bytes)
 
         # extract author from package.json
         author = extract_npm_package_author(package_json)
diff --git a/swh/loader/npm/tests/common.py b/swh/loader/npm/tests/common.py
--- a/swh/loader/npm/tests/common.py
+++ b/swh/loader/npm/tests/common.py
@@ -3,11 +3,10 @@
 # License: GNU General Public License version 3, or any later version
 # See top-level LICENSE file for more information
 
-import json
 import os
 import os.path
 
-import chardet
+from swh.loader.npm.utils import load_json
 
 RESOURCES_PATH = os.path.join(os.path.dirname(__file__), 'resources')
 
@@ -49,8 +48,7 @@
 
     with open(package_metadata_filepath, 'rb') as json_file:
         json_file_bytes = json_file.read()
-        file_encoding = chardet.detect(json_file_bytes)['encoding']
-        package_metadata = json.loads(json_file_bytes.decode(file_encoding))
+        package_metadata = load_json(json_file_bytes)
 
     m.register_uri('GET', package_metadata_url, json=package_metadata)
 
diff --git a/swh/loader/npm/utils.py b/swh/loader/npm/utils.py
--- a/swh/loader/npm/utils.py
+++ b/swh/loader/npm/utils.py
@@ -3,8 +3,13 @@
 # License: GNU General Public License version 3, or any later version
 # See top-level LICENSE file for more information
 
+import json
 import re
 
+from codecs import BOM_UTF8
+
+import chardet
+
 _EMPTY_AUTHOR = {'fullname': b'', 'name': None, 'email': None}
 
 # https://github.com/jonschlinkert/author-regex
@@ -120,3 +125,40 @@
         email = email.encode('utf-8')
 
     return {'fullname': fullname, 'name': name, 'email': email}
+
+
+def _lstrip_bom(s, bom=BOM_UTF8):
+    if s.startswith(bom):
+        return s[len(bom):]
+    else:
+        return s
+
+
+def load_json(json_bytes):
+    """
+    Try to load JSON from bytes and return a dictionary.
+
+    First try to decode from utf-8. If the decoding failed,
+    try to detect the encoding and decode again with replace
+    error handling.
+
+    If JSON is malformed, an empty dictionary will be returned.
+
+    Args:
+        json_bytes (bytes): binary content of a JSON file
+
+    Returns:
+        dict: JSON data loaded in a dictionary
+    """
+    json_data = {}
+    try:
+        json_str = _lstrip_bom(json_bytes).decode('utf-8')
+    except UnicodeDecodeError:
+        encoding = chardet.detect(json_bytes)['encoding']
+        if encoding:
+            json_str = json_bytes.decode(encoding, 'replace')
+    try:
+        json_data = json.loads(json_str)
+    except json.decoder.JSONDecodeError:
+        pass
+    return json_data