diff --git a/jedi/_compatibility.py b/jedi/_compatibility.py
index f06327be..3b3e24fb 100644
--- a/jedi/_compatibility.py
+++ b/jedi/_compatibility.py
@@ -134,13 +134,6 @@ class ImplicitNSInfo(object):
         self.paths = paths
 
 
-# unicode function
-try:
-    unicode = unicode
-except NameError:
-    unicode = str
-
-
 try:
     encoding = sys.stdout.encoding
     if encoding is None:
@@ -156,7 +149,7 @@ def u(string, errors='strict'):
     unicode, because we check that in the beginning).
     """
     if isinstance(string, bytes):
-        return unicode(string, encoding='UTF-8', errors=errors)
+        return str(string, encoding='UTF-8', errors=errors)
     return string
 
 
diff --git a/jedi/api/__init__.py b/jedi/api/__init__.py
index 1bf4e756..2c9b3ef6 100644
--- a/jedi/api/__init__.py
+++ b/jedi/api/__init__.py
@@ -97,9 +97,6 @@ class Script(object):
     :param path: The path of the file in the file system, or ``''`` if
         it hasn't been saved yet.
     :type path: str or None
-    :param encoding: Deprecated, cast to unicode yourself. The encoding of
-        ``code``, if it is not a ``unicode`` object (default ``'utf-8'``).
-    :type encoding: str
     :param sys_path: Deprecated, use the project parameter.
     :type sys_path: typing.List[str]
     :param Environment environment: Provide a predefined :ref:`Environment <environments>`
@@ -115,15 +112,6 @@ class Script(object):
         # An empty path (also empty string) should always result in no path.
         self.path = os.path.abspath(path) if path else None
 
-        if encoding is None:
-            encoding = 'utf-8'
-        else:
-            warnings.warn(
-                "Deprecated since version 0.17.0. You should cast to valid "
-                "unicode yourself, especially if you are not using utf-8.",
-                DeprecationWarning,
-                stacklevel=2
-            )
         if line is not None:
             warnings.warn(
                 "Providing the line is now done in the functions themselves "
diff --git a/jedi/api/classes.py b/jedi/api/classes.py
index 0ba7c3b7..622ec460 100644
--- a/jedi/api/classes.py
+++ b/jedi/api/classes.py
@@ -160,7 +160,7 @@ class BaseName(object):
 
         Finally, here is what you can get from :attr:`type`:
 
-        >>> defs = [str(d.type) for d in defs]  # It's unicode and in Py2 has u before it.
+        >>> defs = [d.type for d in defs]
         >>> defs[0]
         'module'
         >>> defs[1]
diff --git a/jedi/api/keywords.py b/jedi/api/keywords.py
index 999e5dc2..df216f98 100644
--- a/jedi/api/keywords.py
+++ b/jedi/api/keywords.py
@@ -21,9 +21,6 @@ def imitate_pydoc(string):
     if pydoc_topics is None:
         return ''
 
-    # str needed because of possible unicode stuff in py2k (pydoc doesn't work
-    # with unicode strings)
-    string = str(string)
     h = pydoc.help
     with suppress(KeyError):
         # try to access symbols
diff --git a/jedi/api/strings.py b/jedi/api/strings.py
index 6653671e..075f10cc 100644
--- a/jedi/api/strings.py
+++ b/jedi/api/strings.py
@@ -9,7 +9,6 @@ names in a module, but pretty much an arbitrary string.
 """
 import re
 
-from jedi._compatibility import unicode
 from jedi.inference.names import AbstractArbitraryName
 from jedi.inference.helpers import infer_call_of_leaf
 from jedi.api.classes import Completion
@@ -65,7 +64,7 @@ def _completions_for_dicts(inference_state, dicts, literal_string, cut_end_quote
 
 
 def _create_repr_string(literal_string, dict_key):
-    if not isinstance(dict_key, (unicode, bytes)) or not literal_string:
+    if not isinstance(dict_key, (str, bytes)) or not literal_string:
         return repr(dict_key)
 
     r = repr(dict_key)
diff --git a/jedi/inference/analysis.py b/jedi/inference/analysis.py
index df76bf82..1e51db55 100644
--- a/jedi/inference/analysis.py
+++ b/jedi/inference/analysis.py
@@ -49,13 +49,10 @@ class Error(object):
         first = self.__class__.__name__[0]
         return first + str(CODES[self.name][0])
 
-    def __unicode__(self):
+    def __str__(self):
         return '%s:%s:%s: %s %s' % (self.path, self.line, self.column,
                                     self.code, self.message)
 
-    def __str__(self):
-        return self.__unicode__()
-
     def __eq__(self, other):
         return (self.path == other.path and self.name == other.name
                 and self._start_pos == other._start_pos)
diff --git a/jedi/inference/base_value.py b/jedi/inference/base_value.py
index e51f3e66..01c4e7d1 100644
--- a/jedi/inference/base_value.py
+++ b/jedi/inference/base_value.py
@@ -13,7 +13,6 @@ from itertools import zip_longest
 from parso.python.tree import Name
 
 from jedi import debug
-from jedi._compatibility import unicode
 from jedi.parser_utils import clean_scope_docstring
 from jedi.inference.helpers import SimpleGetItemNotFound
 from jedi.inference.utils import safe_property
@@ -384,7 +383,7 @@ def _getitem(value, index_values, contextualized_node):
     unused_values = set()
     for index_value in index_values:
         index = index_value.get_safe_value(default=None)
-        if type(index) in (float, int, str, unicode, slice, bytes):
+        if type(index) in (float, int, str, slice, bytes):
             try:
                 result |= value.py__simple_getitem__(index)
                 continue
diff --git a/jedi/inference/compiled/__init__.py b/jedi/inference/compiled/__init__.py
index b5d435fc..ecb03372 100644
--- a/jedi/inference/compiled/__init__.py
+++ b/jedi/inference/compiled/__init__.py
@@ -1,4 +1,3 @@
-from jedi._compatibility import unicode
 from jedi.inference.compiled.value import CompiledValue, CompiledName, \
     CompiledValueFilter, CompiledValueName, create_from_access_path
 from jedi.inference.base_value import LazyValueWrapper
@@ -45,7 +44,7 @@ def create_simple_object(inference_state, obj):
     Only allows creations of objects that are easily picklable across Python
     versions.
     """
-    assert type(obj) in (int, float, str, bytes, unicode, slice, complex, bool), obj
+    assert type(obj) in (int, float, str, bytes, slice, complex, bool), obj
     compiled_value = create_from_access_path(
         inference_state,
         inference_state.compiled_subprocess.create_simple_object(obj)
diff --git a/jedi/inference/compiled/access.py b/jedi/inference/compiled/access.py
index ce524ff1..1482cabb 100644
--- a/jedi/inference/compiled/access.py
+++ b/jedi/inference/compiled/access.py
@@ -7,10 +7,9 @@ import warnings
 import re
 import builtins
 
-from jedi._compatibility import unicode
 from jedi.inference.compiled.getattr_static import getattr_static
 
-ALLOWED_GETITEM_TYPES = (str, list, tuple, unicode, bytes, bytearray, dict)
+ALLOWED_GETITEM_TYPES = (str, list, tuple, bytes, bytearray, dict)
 
 MethodDescriptorType = type(str.replace)
 # These are not considered classes and access is granted even though they have
@@ -250,7 +249,7 @@ class DirectObjectAccess(object):
         # Avoid some weird hacks that would just fail, because they cannot be
         # used by pickle.
         if not isinstance(paths, list) \
-                or not all(isinstance(p, (bytes, unicode)) for p in paths):
+                or not all(isinstance(p, str) for p in paths):
             return None
         return paths
 
@@ -383,7 +382,7 @@ class DirectObjectAccess(object):
         return [self._create_access(module), access]
 
     def get_safe_value(self):
-        if type(self._obj) in (bool, bytes, float, int, str, unicode, slice) or self._obj is None:
+        if type(self._obj) in (bool, bytes, float, int, str, slice) or self._obj is None:
             return self._obj
         raise ValueError("Object is type %s and not simple" % type(self._obj))
 
diff --git a/jedi/inference/gradual/type_var.py b/jedi/inference/gradual/type_var.py
index 2d508ed8..445fb2ad 100644
--- a/jedi/inference/gradual/type_var.py
+++ b/jedi/inference/gradual/type_var.py
@@ -1,4 +1,3 @@
-from jedi._compatibility import unicode
 from jedi import debug
 from jedi.inference.base_value import ValueSet, NO_VALUES, ValueWrapper
 from jedi.inference.gradual.base import BaseTypingValue
@@ -40,7 +39,7 @@ class TypeVarClass(BaseTypingValue):
             return None
         else:
             safe_value = method(default=None)
-            if isinstance(safe_value, (str, unicode)):
+            if isinstance(safe_value, str):
                 return safe_value
             return None
 
diff --git a/jedi/inference/gradual/typing.py b/jedi/inference/gradual/typing.py
index 94f87ede..2d8a0cb0 100644
--- a/jedi/inference/gradual/typing.py
+++ b/jedi/inference/gradual/typing.py
@@ -7,7 +7,6 @@ This file deals with all the typing.py cases.
 """
 import itertools
 
-from jedi._compatibility import unicode
 from jedi import debug
 from jedi.inference.compiled import builtin_from_name, create_simple_object
 from jedi.inference.base_value import ValueSet, NO_VALUES, Value, \
@@ -459,7 +458,7 @@ class TypedDict(LazyValueWrapper):
         return ValueName(self, self.tree_node.name)
 
     def py__simple_getitem__(self, index):
-        if isinstance(index, unicode):
+        if isinstance(index, str):
             return ValueSet.from_sets(
                 name.infer()
                 for filter in self._definition_class.get_filters(is_instance=True)
diff --git a/jedi/inference/helpers.py b/jedi/inference/helpers.py
index 8f9fa7c8..32650ba8 100644
--- a/jedi/inference/helpers.py
+++ b/jedi/inference/helpers.py
@@ -7,8 +7,6 @@ from contextlib import contextmanager
 
 from parso.python import tree
 
-from jedi._compatibility import unicode
-
 
 def is_stdlib_path(path):
     # Python standard library paths look like this:
diff --git a/jedi/inference/syntax_tree.py b/jedi/inference/syntax_tree.py
index af08a940..a7565e41 100644
--- a/jedi/inference/syntax_tree.py
+++ b/jedi/inference/syntax_tree.py
@@ -5,7 +5,6 @@ import copy
 
 from parso.python import tree
 
-from jedi._compatibility import unicode
 from jedi import debug
 from jedi import parser_utils
 from jedi.inference.base_value import ValueSet, NO_VALUES, ContextualizedNode, \
@@ -587,7 +586,7 @@ def _get_tuple_ints(value):
 def _infer_comparison_part(inference_state, context, left, operator, right):
     l_is_num = is_number(left)
     r_is_num = is_number(right)
-    if isinstance(operator, unicode):
+    if isinstance(operator, str):
         str_operator = operator
     else:
         str_operator = str(operator.value)
diff --git a/jedi/inference/sys_path.py b/jedi/inference/sys_path.py
index d16529c2..fd27bff0 100644
--- a/jedi/inference/sys_path.py
+++ b/jedi/inference/sys_path.py
@@ -2,7 +2,6 @@ import os
 import re
 from importlib.machinery import all_suffixes
 
-from jedi._compatibility import unicode
 from jedi.inference.cache import inference_state_method_cache
 from jedi.inference.base_value import ContextualizedNode
 from jedi.inference.helpers import is_string, get_str_or_none
@@ -203,7 +202,7 @@ def _get_buildout_script_paths(search_path):
         except (UnicodeDecodeError, IOError) as e:
             # Probably a binary file; permission error or race cond. because
             # file got deleted. Ignore it.
-            debug.warning(unicode(e))
+            debug.warning(e)
             continue
 
 
diff --git a/test/test_deprecation.py b/test/test_deprecation.py
index 5c4425c7..ad698250 100644
--- a/test/test_deprecation.py
+++ b/test/test_deprecation.py
@@ -3,8 +3,6 @@ import warnings
 
 import pytest
 
-from jedi._compatibility import u
-
 
 @pytest.fixture(autouse=True)
 def check_for_warning(recwarn):
@@ -40,9 +38,3 @@ def test_usages(Script):
 def test_call_signatures(Script):
     d1, = Script('abs(float(\nstr(', line=1, column=4).call_signatures()
     assert d1.name == 'abs'
-
-
-def test_encoding_parameter(Script):
-    name = u('hö')
-    s = Script(name.encode('latin-1'), encoding='latin-1')
-    assert s._module_node.get_code() == name