diff --git a/doc/progress.rst b/doc/progress.rst
index dac22ff22..c6ce7f30e 100644
--- a/doc/progress.rst
+++ b/doc/progress.rst
@@ -20,6 +20,8 @@ Changelog
   OpenML.
 * ADD #564: New helpers to access the structure of a flow (and find its
   subflows).
+* ADD #618: The software will from now on retry to connect to the server if a
+  connection failed. The number of retries can be configured.
 * FIX #538: Support loading clustering tasks.
 * FIX #464: Fixes a bug related to listing functions (returns correct listing
   size).
diff --git a/openml/_api_calls.py b/openml/_api_calls.py
index 6a1086221..707516651 100644
--- a/openml/_api_calls.py
+++ b/openml/_api_calls.py
@@ -1,9 +1,7 @@
-import io
-import os
+import time
 import requests
 import warnings
 
-import arff
 import xmltodict
 
 from . import config
@@ -11,12 +9,9 @@
                          OpenMLServerNoResult)
 
 
-def _perform_api_call(call, data=None, file_elements=None,
-                      add_authentication=True):
+def _perform_api_call(call, data=None, file_elements=None):
     """
     Perform an API call at the OpenML server.
-    return self._read_url(url, data=data, filePath=filePath,
-    def _read_url(self, url, add_authentication=False, data=None, filePath=None):
 
     Parameters
     ----------
@@ -27,8 +22,6 @@ def _read_url(self, url, add_authentication=False, data=None, filePath=None):
     file_elements : dict
         Mapping of {filename: str} of strings which should be uploaded as
         files to the server.
-    add_authentication : bool
-        Whether to add authentication (api key) to the request.
 
     Returns
     -------
@@ -50,12 +43,12 @@ def _read_url(self, url, add_authentication=False, data=None, filePath=None):
 
 
 def _file_id_to_url(file_id, filename=None):
-    '''
+    """
      Presents the URL how to download a given file id
      filename is optional
-    '''
+    """
     openml_url = config.server.split('/api/')
-    url = openml_url[0] + '/data/download/%s' %file_id
+    url = openml_url[0] + '/data/download/%s' % file_id
     if filename is not None:
         url += '/' + filename
     return url
@@ -71,7 +64,12 @@ def _read_url_files(url, data=None, file_elements=None):
         file_elements = {}
     # Using requests.post sets header 'Accept-encoding' automatically to
     # 'gzip,deflate'
-    response = requests.post(url, data=data, files=file_elements)
+    response = send_request(
+        request_method='post',
+        url=url,
+        data=data,
+        files=file_elements,
+    )
     if response.status_code != 200:
         raise _parse_server_exception(response, url=url)
     if 'Content-Encoding' not in response.headers or \
@@ -87,12 +85,16 @@ def _read_url(url, data=None):
         data['api_key'] = config.apikey
 
     if len(data) == 0 or (len(data) == 1 and 'api_key' in data):
-        # do a GET
-        response = requests.get(url, params=data)
-    else: # an actual post request
+        response = send_request(
+            request_method='get', url=url, data=data,
+        )
+
+    else:
         # Using requests.post sets header 'Accept-encoding' automatically to
         #  'gzip,deflate'
-        response = requests.post(url, data=data)
+        response = send_request(
+            request_method='post', url=url, data=data,
+        )
 
     if response.status_code != 200:
         raise _parse_server_exception(response, url=url)
@@ -102,12 +104,44 @@ def _read_url(url, data=None):
     return response.text
 
 
+def send_request(
+    request_method,
+    url,
+    data,
+    files=None,
+):
+    n_retries = config.connection_n_retries
+    response = None
+    with requests.Session() as session:
+        # Start at one to have a non-zero multiplier for the sleep
+        for i in range(1, n_retries + 1):
+            try:
+                if request_method == 'get':
+                    response = session.get(url, params=data)
+                elif request_method == 'post':
+                    response = session.post(url, data=data, files=files)
+                else:
+                    raise NotImplementedError()
+                break
+            except (
+                    requests.exceptions.ConnectionError,
+                    requests.exceptions.SSLError,
+            ) as e:
+                if i == n_retries:
+                    raise e
+                else:
+                    time.sleep(0.1 * i)
+    if response is None:
+        raise ValueError('This should never happen!')
+    return response
+
+
 def _parse_server_exception(response, url=None):
     # OpenML has a sopisticated error system
     # where information about failures is provided. try to parse this
     try:
         server_exception = xmltodict.parse(response.text)
-    except:
+    except Exception:
         raise OpenMLServerError(('Unexpected server error. Please '
                                  'contact the developers!\nStatus code: '
                                  '%d\n' % response.status_code) + response.text)
@@ -117,7 +151,7 @@ def _parse_server_exception(response, url=None):
     additional = None
     if 'oml:additional_information' in server_exception['oml:error']:
         additional = server_exception['oml:error']['oml:additional_information']
-    if code in [372, 512, 500, 482, 542, 674]: # datasets,
+    if code in [372, 512, 500, 482, 542, 674]:
         # 512 for runs, 372 for datasets, 500 for flows
         # 482 for tasks, 542 for evaluations, 674 for setups
         return OpenMLServerNoResult(code, message, additional)
diff --git a/openml/config.py b/openml/config.py
index 897eadd2b..0ca5936a0 100644
--- a/openml/config.py
+++ b/openml/config.py
@@ -21,6 +21,7 @@
     'verbosity': 0,
     'cachedir': os.path.expanduser(os.path.join('~', '.openml', 'cache')),
     'avoid_duplicate_runs': 'True',
+    'connection_n_retries': 2,
 }
 
 config_file = os.path.expanduser(os.path.join('~', '.openml' 'config'))
@@ -32,6 +33,9 @@
 # The current cache directory (without the server name)
 cache_directory = ""
 
+# Number of retries if the connection breaks
+connection_n_retries = 2
+
 
 def _setup():
     """Setup openml package. Called on first import.
@@ -46,6 +50,7 @@ def _setup():
     global server
     global cache_directory
     global avoid_duplicate_runs
+    global connection_n_retries
     # read config file, create cache directory
     try:
         os.mkdir(os.path.expanduser(os.path.join('~', '.openml')))
@@ -57,6 +62,12 @@ def _setup():
     server = config.get('FAKE_SECTION', 'server')
     cache_directory = os.path.expanduser(config.get('FAKE_SECTION', 'cachedir'))
     avoid_duplicate_runs = config.getboolean('FAKE_SECTION', 'avoid_duplicate_runs')
+    connection_n_retries = config.get('FAKE_SECTION', 'connection_n_retries')
+    if connection_n_retries > 20:
+        raise ValueError(
+            'A higher number of retries than 20 is not allowed to keep the '
+            'server load reasonable'
+        )
 
 
 def _parse_config():
diff --git a/openml/testing.py b/openml/testing.py
index 80c4b3183..586345a9c 100644
--- a/openml/testing.py
+++ b/openml/testing.py
@@ -65,6 +65,10 @@ def setUp(self):
                 with open(openml.config.config_file, 'w') as fh:
                     fh.write('apikey = %s' % openml.config.apikey)
 
+        # Increase the number of retries to avoid spurios server failures
+        self.connection_n_retries = openml.config.connection_n_retries
+        openml.config.connection_n_retries = 10
+
     def tearDown(self):
         os.chdir(self.cwd)
         try:
@@ -76,6 +80,7 @@ def tearDown(self):
             else:
                 raise
         openml.config.server = self.production_server
+        openml.config.connection_n_retries = self.connection_n_retries
 
     def _get_sentinel(self, sentinel=None):
         if sentinel is None:
diff --git a/tests/test_runs/test_run_functions.py b/tests/test_runs/test_run_functions.py
index 16e433979..8c542e39b 100644
--- a/tests/test_runs/test_run_functions.py
+++ b/tests/test_runs/test_run_functions.py
@@ -735,17 +735,20 @@ def test_get_run_trace(self):
             if 'Run already exists in server' not in e.message:
                 # in this case the error was not the one we expected
                 raise e
-            # run was already
-            flow = openml.flows.sklearn_to_flow(clf)
-            flow_exists = openml.flows.flow_exists(flow.name, flow.external_version)
-            self.assertIsInstance(flow_exists, int)
-            self.assertGreater(flow_exists, 0)
-            downloaded_flow = openml.flows.get_flow(flow_exists,
-                                                    reinstantiate=True)
-            setup_exists = openml.setups.setup_exists(downloaded_flow)
-            self.assertIsInstance(setup_exists, int)
-            self.assertGreater(setup_exists, 0)
-            run_ids = _run_exists(task.task_id, setup_exists)
+            # run was already performed
+            message = e.message
+            if sys.version_info[0] == 2:
+                # Parse a string like:
+                # 'Run already exists in server. Run id(s): set([37501])'
+                run_ids = (
+                    message.split('[')[1].replace(']', '').
+                    replace(')', '').split(',')
+                )
+            else:
+                # Parse a string like:
+                # "Run already exists in server. Run id(s): {36980}"
+                run_ids = message.split('{')[1].replace('}', '').split(',')
+            run_ids = [int(run_id) for run_id in run_ids]
             self.assertGreater(len(run_ids), 0)
             run_id = random.choice(list(run_ids))